How to Run: Brand Awareness Survey

Introduction

A brand awareness survey measures what share of a defined population knows your brand, compared with named competitors in the same sample. Survey category buyers rather than your own users, ask unaided recall before aided recognition, include two fictitious foil brands, size the sample for the precision your decision needs, and compare brands with a paired test. Skipping the foils or the paired test can produce a misleading number.

Awareness numbers show up in board decks, launch plans and rebrand debates, and most of them were built on a sample that couldn’t support the claim. The errors that matter are made before fieldwork starts and can’t be repaired at analysis. A sample drawn from your own users overstates market awareness, and comparing two brands as though they came from separate samples can hide a real lead.

Last updated October 2026.

What a Brand Awareness Survey Is

A brand awareness survey is a cross-sectional questionnaire that estimates how many people in a defined population can recall or recognize a brand. It produces proportions, not scores, and each proportion only means something alongside the population it was measured in.

The Three Measures

Awareness is typically reported at three levels of difficulty: unaided recall, top-of-mind awareness and aided recognition.

Unaided recall asks respondents to name brands in a category with no list shown. Top-of-mind is the share of respondents naming your brand first. Aided recognition shows a list and asks which brands the respondent has heard of. Measure all three in one survey of category buyers, asking unaided before aided, and compare each figure with competitors measured in the same sample.

The three aren’t independent metrics. Laurent, Kapferer and Roussel (1995) found they’re strongly but nonlinearly related across brands, and Romaniuk, Sharp, Paech and Driesener (2004) found they’re not simply one construct, which the brand tracking guide covers in depth.

How a Single Read Differs From a Tracker

A single read answers a level question. A tracker answers a change question, and the two need different designs. A tracker freezes its instrument across waves, while a single read can optimize its instrument for this one measurement.

The trade is that a single read can never tell you whether you’re growing, which is often the next question a board asks.

What Kind of Evidence It Produces

A brand awareness survey produces descriptive evidence about memory at one point in time. It doesn’t produce causal evidence, attitudinal depth, or evidence about purchase.

That scope is narrower than the way awareness numbers are often presented in decks. A board slide that reads 40 percent aided recognition as 40 percent of the market has asked the number to carry more than it can.

When to Use a Brand Awareness Survey

Use a brand awareness survey when a decision depends on how many people in a specific population know your brand today, relative to specific competitors. The decision should exist before the survey does.

Four Situations That Justify One

Four situations generally justify the effort:

Setting a baseline before a launch, rebrand, or first major campaign
Sizing awareness in a new market, region, or buyer segment before entering it
Answering an investor or board question about category position with evidence
Checking whether a B2B (business-to-business) brand is known among a named buyer role or account list

Each of these is a level question with a population attached. The baseline case is the most common, and it’s also the one most likely to turn into a tracker, which the design chapter addresses first.

The Question It Answers Well

“Among category buyers in this market, what share can recall or recognize our brand, and how does that compare with the competitors they’re most likely to consider?”

That question is answerable with a modest sample and a well-built list. Most of what follows is about answering it without introducing avoidable error.

When a Single Read Is Enough

A single read is frequently enough when the decision is a go or no-go rather than a trend. A team deciding whether to lead a new market with brand spend or with sales outreach typically needs one honest number, not a quarterly series.

The same logic applies to a rebrand. A team choosing between keeping a name and replacing it needs to know how much recognition the current name carries among buyers today, and a single read measures exactly that.

Awareness is also often one block inside a broader usage and attitudes study. If you also need category habits, purchase frequency and attitudes in the same fieldwork, run that study and treat the awareness block as its first section.

When Not to Use a Brand Awareness Survey

A brand awareness survey can’t tell you six things that teams commonly ask it to. Each limitation below names the instrument that answers the question instead.

It Can't Tell You Whether Your Marketing Caused the Level

A single read has no counterfactual. A high awareness score after a campaign may reflect the campaign, prior spend, word of mouth, or simply the brand’s size.

The better instrument is an exposed-versus-control brand lift study or a geographic holdout. Google’s Brand Lift, for example, compares survey responses from people who saw an ad with people who were eligible but didn’t see it, and that comparison supplies the missing counterfactual.

It Can't Tell You Whether Awareness Will Turn Into Choice

Knowing a brand isn’t the same as considering it, and considering it isn’t the same as buying it. Awareness is generally a necessary condition for choice and rarely a sufficient one.

The better instrument is a consideration and category entry point battery, which the brand tracking guide sets out, or a brand perception study when the question is what people think of you.

It Can't Tell You Whether You're Growing

One read is one point. Comparing it with a different survey run last year on a different list and a different sample isn’t a trend, however tempting the chart looks.

The better instrument is the same questionnaire fielded repeatedly under constant conditions, which is a tracker. The brand tracking guide covers the wave design and change tests a tracker needs.

It Can't Tell You Which Message or Creative Drives Recall

An awareness survey measures an outcome with no manipulation in it. It can’t isolate which ad, headline or claim put the brand in memory.

The better instrument is monadic message testing, where each respondent sees one variant, or MaxDiff across a set of claims.

It Can't Tell You Why Unaware Buyers Are Unaware

Awareness measures incidence, not reasoning. An open-text probe gets somewhat closer, but a respondent who has never heard of you can’t explain the absence of a memory.

The better instrument is depth interviews with category buyers about how they search, shortlist and decide. Interviews are a qualitative instrument, not a survey tool. They trade incidence for reasoning, drawing depth from a small number of category buyers, and reasoning is what this limitation needs. The guide on when to run an AI-moderated interview instead of a survey covers that choice.

It Can't Tell You What Awareness Is Worth

A brand awareness survey produces percentage points, not revenue. It can’t say whether 10 more points of aided recognition would pay back the media spend needed to buy them.

The better instrument is a market-response or marketing-mix model that relates brand metrics and spend to sales over time. Feed the survey’s figures into the model as one input alongside spend, distribution and price. A model built without a defensible awareness measure is guessing at the brand term, which is the one place a single read contributes to the value question.

What These Limits Have in Common

Every limit above is a question about cause, change, choice, content, reasoning or value. A single awareness read answers none of those, and it answers the level question well.

Pair the awareness read with the instrument that owns the question, and name the paired instrument in the study plan, such as a tracker for the growth question and a lift study for the campaign question.

Study Design, and the Four Forks That Decide Everything Else

Four decisions shape a brand awareness survey, and each should be made before a question is written. Getting any of them wrong usually can’t be repaired at analysis.

Fork 1: A One-Off Read or Wave 1 of a Tracker

If there’s any realistic chance you’ll measure again, build the survey as wave 1 of a tracker. Freeze the brand list, the question wording and the sample definition from the start.

A one-off built without that discipline typically becomes the baseline anyway, six months later, when someone asks whether the campaign worked. At that point a changed list or reworded question breaks the comparison, and nobody can recover it.

Write the decision into the study plan now. If it’s a true one-off, say so, and optimize freely.

Fork 2: The Category Population or Your Own Users

Survey category buyers, not your own users. Awareness among people who already use you is close to universal by construction, and it tells you almost nothing about the market.

The closest peer-reviewed evidence concerns advertising awareness, not brand awareness. Romaniuk and Wight (2009), in the International Journal of Market Research, found brand users were about 2.5 times as likely as non-users to recall advertising exposure on unprompted measures, narrowing to about 1.7 times when the brand was prompted. Applying the same skew to brand recall is an inference, but it points the same way.

For a small or new brand, most of the population is non-users. A survey of your customer list will typically overstate awareness, and the advertising evidence suggests the skew is larger on unprompted measures than on prompted ones.

Category buyers commonly have to be reached through a research panel with screening. That makes incidence, the share of panel members who qualify, a planning input instead of an afterthought.

In Sprig: Response-based quotas can balance a sample by region, role or category usage. The limits: the quota documentation says quotas are only available on the Enterprise plan, quota screeners must be single select, rating scale or NPS (Net Promoter Score) questions, only completed responses count, and criteria can’t change after launch, though targets can be raised or lowered. Settle the screener first.

What Category Incidence Does to Cost

Incidence is the share of panel members who pass the screener. A category bought by 40 percent of adults is cheap to reach, and one bought by 3 percent of adults may need 13 times as many screened entrants per qualified complete.

Get an incidence estimate before committing to a sample size. Where no estimate exists, run a short soft launch and measure the pass rate directly, then size the full study from it.

Fork 3: Which Measures to Field

Field unaided recall first and aided recognition second, and derive top-of-mind from the first unaided mention. That order isn’t a preference. Showing a list first contaminates every recall measure after it.

Which measure leads the interpretation depends on brand size. For a small or new brand, lead with aided recognition, because unaided recall for a small brand often sits in single digits, where intervals are wide and a handful of mentions moves the figure.

For an established brand, unaided recall generally carries more information, because aided recognition is often close to ceiling and differences between large brands shrink there.

Fork 4: Is the Real Question Whether a Campaign Worked

If the real question is whether a campaign raised awareness, a brand awareness survey is the wrong design. A single read, or even a before-and-after pair on the same population, can’t separate the campaign from everything else that happened in between.

Use an exposed-versus-control design instead. Advertising platforms run these as brand lift studies, and a survey can approximate one by sampling an exposed group and a matched unexposed group at the same time. That design has its own sampling and contamination problems, and it deserves more than this section. Treat this fork as a signpost, not a method.

Write the Decisions Into a One-Page Study Plan

Before building anything, write down the answer to each fork in plain sentences. The plan should name the population, the brand list, the measures, whether this is wave 1 of a tracker, and the decision the result will inform.

It stops the questionnaire drifting during review, and it gives the next team a way to repeat the study exactly.

The Design Tree

Read the tree top to bottom and stop at the first branch that applies:

The tree forces the level question and the change question apart before either is confused with the causal question. A study that lands on the first row should stop here and be scoped as a lift study instead.

Writing the Instrument

A brand awareness instrument is short and strictly ordered: a category screener, an unaided recall question in open text, an aided recognition question with your brand, competitors and two foil brands, then optional consideration and demographics. Never show a brand name before the recall question.

Define the Category the Way Buyers Do

The unaided question is only as good as its category cue. “Which project management tools come to mind?” and “Which tools come to mind for planning a team’s work week?” will return different brand sets from the same person.

Use the words a buyer would use when the need arises, not your internal category label. If the category is contested, pilot two cues with a small sample and keep the one that returns the fewest off-category answers.

The Unaided Question

Ask for every brand the respondent can think of, as open text, with no list and no examples.

“When you think of [CATEGORY DESCRIBED IN BUYER LANGUAGE], which brands come to mind? Please list as many as you can think of.”

Use a single open-text box, or a set of numbered boxes if your analysis needs mention order preserved cleanly. The first brand typed is top-of-mind.

Don’t add an “I can’t think of any” checkbox beside the text box. It offers an effortless exit, which invites respondents to skip the effort recall needs. A blank or typed “none” is counted as a non-response and stays in the denominator.

The Aided Question

Show the brand list as a multi-select, randomized, with an exclusive “None of these” pinned to the bottom.

“Which of the following brands have you heard of, even if only by name? Select all that apply.”

The phrase “even if only by name” matters. Without it, some respondents read the question as asking which brands they know well, which measures familiarity instead of recognition.

Building the Brand List

Include your brand, the competitors a buyer is most likely to consider, and two fictitious foil brands.

Keep the real list short enough to scan. Long lists are tiring to scan, and randomization spreads any position effect across brands without removing it.

Choose competitors from buyer behavior, not from your own competitive slide, which often reflects who the company worries about instead of who buyers consider. Sales loss reasons, review-site category pages and the unaided answers from a small pilot are all better sources than an internal opinion.

Foil Brands

Add two invented brand names that sound plausible in the category, and test each with a web search first. Any respondent selecting one is claiming to recognize a brand that doesn’t exist, which shows how much aided recognition is inflated.

Foils apply the over-claiming technique from Paulhus, Harms, Bruce and Lysy (2003) in the Journal of Personality and Social Psychology, applied to brands. Any familiarity claimed for a non-existent item counts as self-enhancement, not knowledge.

Test each foil name with a web search before fielding, since invented names frequently turn out to exist. A foil that turns out to be a real regional company is no longer a foil.

The Instrument Spec

This is the full questionnaire in fielding order, ready to adapt:

The consideration item at position 6 is a bridge to the questions awareness can’t answer. It’s typically the most useful extra item, because the ratio of considering to aware is often more actionable than either number alone.

In Sprig: Multiple choice questions support randomized order, pinned choices, an exclusive “None of the above” on multi-select, and images on each answer choice for logo recognition, per the question types documentation. The limit matters for items 4 to 6: questions with images can’t be a carry forward source or destination, so logo-recognized brands can’t be piped into the consideration question automatically.

Scale and Scoring Choices

Awareness items are generally binary, so there’s no rating scale to choose. The scoring decisions are about denominators, mention order and how to count people who select a foil.

The Denominators

Report every awareness figure as a share of qualified respondents, including those who answered “None of these” or typed nothing. Excluding non-answers from the denominator typically inflates every brand’s score.

State the denominator in every table. “48 percent aware” means nothing until the reader knows it’s 48 percent of 600 screened category buyers in one market.

Top-of-Mind Scoring

Score top-of-mind as the share whose first coded unaided mention is your brand. Only one brand can be first per respondent, so top-of-mind shares across brands sum to the share of respondents who named anything, and never more.

That constraint makes top-of-mind a share metric, not an awareness metric. It’s often the least stable of the three for small brands, because a handful of first mentions moves it several points.

Unaided Scoring

Score unaided recall as the share mentioning the brand anywhere in their answer, after coding. Misspellings, abbreviations and product names all need to map to the right brand before counting.

The coding step is where unaided numbers are usually won or lost. A frame that misses “zen desk” for a brand called Zendesk under-counts that brand, and the error is invisible in the final table.

Aided Scoring and the Foil Rate

Score aided recognition as the share selecting the brand from the list. Then report the foil rate beside it: the share of respondents who selected at least one fictitious brand.

Don’t subtract the foil rate from every score. Report a sensitivity check instead: recompute aided recognition excluding respondents who selected any foil, and show both numbers.

Subtraction assumes over-claiming is uniform across brands, and there’s little reason to expect that. Gold’s 2013 research-on-research in Quirk’s reported some support for lists with no real brands producing more false positives, without publishing the figures, so the inflation may depend on context.

Logo Recognition Scoring

If the survey includes a logo recognition item, score it separately from name recognition and never merge the two. A respondent who recognizes a logo but not the name, or the reverse, is telling you something about which brand asset carries recognition.

Show logos with the brand name removed. A logo containing the wordmark tests name recognition twice instead of visual recognition once.

Consideration as a Conversion Ratio

Report consideration as a share of those aware, not only as a share of everyone. A brand with 30 percent awareness and 60 percent consideration among the aware is in a different position from one with 60 percent awareness and 30 percent consideration among the aware.

Both have 18 percent consideration overall. The first has an awareness problem and the second has a perception problem, and they call for different budgets.

Sample Size and Precision

Size a single-read awareness survey from the margin of error your decision can tolerate: at 50 percent awareness, 385 qualified completes give about plus or minus 5 points at 95 percent confidence, and 1,068 give about plus or minus 3. No methodology body publishes a minimum sample for awareness studies, and this guide doesn’t invent one.

The Arithmetic

For a proportion, the 95 percent margin of error is 1.96 multiplied by the square root of p times one minus p, divided by n. Here p is the true awareness level and n is the number of qualified completes.

Solving for n gives the sample needed for a chosen margin: n equals 1.96 squared, times p times one minus p, divided by the margin squared. The margin is widest at 50 percent awareness, which makes 50 percent the conservative planning assumption when you have no prior estimate.

Derived Table: Margin of Error for a Single Read

The table below is derived from that formula at 95 percent confidence, in percentage points. It’s arithmetic, not a published norm.

| True awareness | 200 completes | 400 completes | 1,000 completes | |:--------------:|:-------------:|:-------------:|:---------------:| | 50% | ±6.9 | ±4.9 | ±3.1 | | 20% | ±5.5 | ±3.9 | ±2.5 | | 10% | ±4.2 | ±2.9 | ±1.9 | | 5% | ±3.0 | ±2.1 | ±1.4 |

Read the table as a trade-off, since precision costs sample. At 400 completes, a brand near 50 percent aided recognition is known to within about five points either way, which is typically enough for a go or no-go decision and rarely enough to rank two close competitors.

Derived Table: Completes Needed for a Target Margin

Working the formula the other way gives the qualified completes needed:

| Expected awareness | Target margin | Completes needed | |:------------------:|:-------------:|:----------------:| | 50% | ±5 points | 385 | | 50% | ±4 points | 601 | | 50% | ±3 points | 1,068 | | 20% | ±3 points | 683 | | 10% | ±2 points | 865 |

Low Unaided Scores Need a Different Interval

Unaided recall for a small brand is often in single digits, and the simple formula behaves badly there. At 5 percent awareness and 200 completes, the simple interval runs from 2.0 to 8.0 percent, symmetric around the estimate.

The Wilson score interval runs from 2.7 to 9.0 percent, asymmetric because awareness can’t fall below zero. Use the Wilson interval for every awareness figure, since it behaves well at low and high proportions, and name the method in the report.

Segments Divide the Sample

Every segment you intend to report needs its own precision. A 600-complete study split across three regions is three studies of about 200, each with a margin near seven points at 50 percent awareness.

Decide the reporting cuts before sizing. Teams commonly size for the total and then report the segments, which is where confident claims about small differences usually come from.

Weighting Costs Precision

If the achieved sample is weighted to match a population profile, the effective sample size shrinks. Kish’s design effect for weighting is one plus the squared coefficient of variation of the weights.

Weights with a coefficient of variation of 0.5 give a design effect of 1.25, turning 600 completes into an effective 480. Plan the sample on the effective number, not the raw one.

Small Populations Change the Arithmetic

When the population itself is small, the finite population correction applies. A B2B study of an account list with 3,000 qualified buyers is sampling a meaningful share of the whole population, and the margin shrinks accordingly.

Multiply the margin by the square root of one minus n divided by the population size. For 400 completes from 3,000 buyers, that factor is about 0.93, so the plus or minus 4.9 points at 50 percent becomes about 4.6. The correction only matters once the sample is more than a few percent of the population.

Comparing Two Brands Needs Its Own Calculation

The margins above describe one brand’s score. Comparing your brand with a competitor in the same sample is a different calculation, covered in the analysis chapter, and it’s often more precise than two separate margins suggest.

Audience, Targeting and Screening

The population for a brand awareness survey is category buyers in a defined market, qualified by behavior instead of self-description. Every awareness figure inherits its meaning from that definition.

Define the Population in One Sentence

Write the population as a sentence before building any screener. “Adults in the United States who bought or renewed a meal-kit subscription in the past 12 months” is a population. “Consumers interested in food” isn’t.

The sentence should name a market, a category behavior and a time window. Each of those three will become a screener question.

Screen on Behavior, Not Identity

Ask what respondents did, not what they are, because behavior is often easier to verify than identity. “Have you personally purchased or renewed any of the following in the past 12 months?” screens more reliably than “Are you a meal-kit customer?”, which invites a yes from anyone who ever tried one.

Hide the qualifying answer among plausible distractors. A screener that makes the right answer obvious commonly qualifies people who guessed it.

B2B Screening

For a B2B awareness survey, screen on role and decision involvement, not only job title. A procurement analyst and an end user at the same company may know entirely different vendor sets.

Ask about involvement in a recent purchase decision for the category, and qualify only those who influenced or made one. Titles vary so much across companies that a title filter alone generally lets in the wrong people.

Quotas, and When to Use Them

Set quotas when the population has a known profile and the panel’s natural response pattern would skew away from it. Region, age band and company size are common quota variables.

Set them on the one or two variables most likely to move awareness, not on every variable you can think of. Every quota raises cost and slows fielding.

Report Incidence Alongside the Results

Record how many respondents entered, how many were screened out at each question, and how many qualified. That funnel is part of the finding, since the share of a panel that buys the category is itself market information.

In Sprig: Panel recruitment offers over 300 filters, including location, age and job title, and shows a live count of matching participants. The limits: panels are available to Enterprise teams on link surveys only, need credits bought through a Sprig contact, and can’t be edited after launch beyond a few controls such as participant numbers, so the population sentence has to be right first.

Fielding and Delivery

Field a brand awareness survey to a panel or a purchased list of category buyers, over a short window, in a single mode. Mixing channels or stretching the field period introduces differences the analysis can’t separate from awareness.

Panel, List or Own Channels

A research panel is typically the only practical route to a market-level awareness number. Your own email list, website visitors and in-product audiences are all populations that already know you.

Own channels still have a legitimate use. Awareness of a new product line among existing customers is a real question, and an in-product or email survey answers it well, provided the report names the population.

Keep the Field Window Short

Field over days instead of weeks where incidence allows. A long field period risks a news event, a competitor launch or your own campaign landing mid-study, and splitting the data afterwards leaves each part with half the sample, which widens each margin by about 40 percent.

One Mode Throughout

Hold the delivery mode constant wherever possible, since mode effects are frequently invisible in the data. A survey answered on a phone and the same survey answered on a desktop can present a long brand list differently, and randomization only partly protects against layout effects.

If mixed devices are unavoidable, record the device and check whether aided recognition differs by device before pooling.

Keep the Brand Out of the Invitation

The survey title, invitation email and panel listing can’t name your brand or any brand on the list. A title such as “Tell us about Kestrel” primes recall before the first question is seen.

Describe the survey by its category instead, and keep the sponsor anonymous. Respondents who know the sponsor can over-report awareness of that brand, which is precisely the number the study exists to measure.

Run a Soft Launch

Release the survey to a small share of the target sample first, then stop and check. Read the unaided answers, confirm the screener pass rate, and confirm the aided list displays correctly on phones.

A soft launch commonly catches a screener that leaks or a category cue that returns off-category answers. Fixing either after full fielding means discarding data, not editing a question.

Timing a Brand Awareness Survey

Time a single read to the decision it serves, and avoid fielding during your own marketing activity unless that activity is the thing being measured. A baseline fielded the week a campaign launches is no longer a baseline.

Baselines Go Before Anything Visible

For a pre-launch or pre-campaign baseline, field and close before any public activity starts. That includes press, paid social tests and sales outreach at scale, all of which commonly leak into category conversations sooner than expected.

Seasonality Matters for Category Screening

Category incidence often moves with the calendar. A survey of recent tax-software buyers fielded in April reaches a different population from the same survey fielded in October.

Note the field dates beside every figure. If the survey later becomes wave 1 of a tracker, the next wave should be fielded in the same season.

Competitor Events

Check competitors' announced launches and campaigns before setting field dates. A major competitor launch during fieldwork can depress your share of unaided mentions without any change in your own awareness, because respondents typically name only a handful of brands unprompted.

Holidays and News Cycles

Avoid fielding across major holidays, when panel composition can shift toward people with more free time. A category-relevant news story during fieldwork can also raise unaided mentions for the brands named in it, so note any such event beside the field dates.

Quality Control and Data Hygiene

Four checks target the failure modes specific to awareness data: over-claiming, list straight-lining, junk unaided answers and a leaking screener. Run them before any number reaches a slide.

Check the Foil Rate First

Look at the share of respondents who selected at least one foil brand, which is often the fastest check available. It’s the most direct indicator of how much aided recognition is inflated by over-claiming or inattention.

Practitioner evidence suggests the rate isn’t trivial. In a research-on-research study reported in Quirk’s in 2013, Peter Gold reported that two fictitious brands in an artisan-bread study drew 9 and 12 percent awareness, and that 9 to 13 percent of respondents claimed awareness of four or more fictitious brands across the categories tested.

That’s trade press, not peer-reviewed evidence, and the figures come from one research firm’s studies. Treat them as a reason to measure your own foil rate, not as an expected value.

Check for Straight-Lining on the Aided List

Flag respondents who selected every brand on the list, foils included. Selecting everything is a pattern, not an answer, and it inflates every selected brand, which the foils should reveal.

Decide in advance how you’ll handle them, and apply the rule before looking at brand results. Choosing an exclusion rule after seeing which brands it helps is a common way to bias a result without meaning to.

Check the Unaided Answers for Junk

Read a sample of unaided responses before coding. Keyboard mashing, pasted text, and answers that name the survey platform or the panel are all signs of a respondent who wasn’t engaged.

Count them, report the count, and remove them from both unaided and aided analysis. A respondent who couldn’t produce a real unaided answer is unlikely to have produced a careful aided one.

Check the Screener Pass Rate Against Expectations

Compare the share of respondents who qualified against what you expected. A pass rate far above plausible category incidence usually means the screener is leaking, and the resulting sample isn’t the population you defined.

In Sprig: Bot detection adds an invisible reCAPTCHA to survey links and saves a human-likelihood score from 0.0 to 1.0 with each response so suspected bots can be excluded. The limit is that the documentation publishes no accuracy or false-positive rate for the score, so foil brands and the checks above are still needed to judge human over-claiming.

Analysis: The Core Calculation

The core calculation is a set of proportions with intervals, a foil sensitivity check, and paired comparisons between brands measured on the same respondents. The paired comparison is the step most analyses get wrong.

Step 1: Compute Each Brand's Proportions

For each brand, compute unaided recall, top-of-mind and aided recognition as shares of all qualified respondents. Put a 95 percent interval around each, using the Wilson method.

Step 2: Run the Foil Sensitivity Check

Recompute aided recognition excluding respondents who selected any foil. Report both figures side by side.

If a brand’s aided score barely moves, over-claiming isn’t driving it. If it drops, report the size of the drop in points beside the brand’s interval and say the headline figure carries some inflation. The two figures share most of their respondents, so their margins can’t be compared as though they were independent.

Step 3: Compare Brands With a Paired Test

Your brand’s awareness and a competitor’s awareness come from the same respondents, so they aren’t independent samples. Wild and Seber (1993), in The American Statistician 47, pages 178 to 181, set out why comparing two proportions from the same survey needs a different standard error from the familiar two-sample test.

For two brands measured on the same people, the variance of the difference depends on the respondents who know one brand but not the other. Those discordant respondents carry all the information about the gap.

The standard paired formula is: variance of the difference equals the share aware of A only, plus the share aware of B only, minus the squared difference between the two shares, all divided by n. The standard error is the square root of that.

A Worked Example

Take 600 qualified respondents. Kestrel, an invented brand for this example, has 45 percent aided recognition, and its competitor Plinth has 41 percent. Thirty-three percent recognize both.

So 12 percent know only Kestrel and 8 percent know only Plinth. The paired variance is (0.12 plus 0.08 minus 0.0016, the squared four-point gap) divided by 600, which gives a standard error of 1.8 points. The 95 percent interval for the four-point gap runs from 0.4 to 7.6 points. Zero is excluded, so Kestrel’s lead is distinguishable from noise at this sample size.

Now run the two-sample test that commonly gets used instead. It treats the two scores as separate samples, gives a standard error of 2.9 points, and produces an interval from minus 1.6 to 9.6 points. That interval includes zero. The wrong test would have told you the lead wasn’t real. These numbers were chosen to show the contrast, so a real gap won’t always flip this way.

Why the Paired Test Is Usually More Precise

Awareness of brands in the same category is typically positively correlated. People engaged with a category tend to know several brands in it, so the scores move together across respondents.

The two-sample formula ignores that correlation and overstates the uncertainty of the gap. Compute both and report the paired one.

Step 4: Compute Consideration Among the Aware

For each brand, divide the share considering by the share aware. Put an interval on the ratio using the aware respondents as the denominator, since those are the only people who could have considered it.

Step 5: Record Every Analytic Decision

Write down the exclusion rules, the code frame version, the interval method and the paired formula used. A single read that becomes wave 1 of a tracker later is only comparable if the next analyst can reproduce every one of those choices.

Benchmarks and What a Good Awareness Result Looks Like

There’s no sourced cross-industry benchmark for a good brand awareness score. In a single-read study, the only defensible comparison is your competitors, measured in the same sample, on the same list, at the same time.

Why No Benchmark Transfers

The brand tracking guide sets out the sourced reasons in its benchmarks section, and they apply equally to a single read. Awareness scales with brand size, the three measures are non-linearly related, and the category cue changes the answer. This section covers what to do instead.

Your Competitor Set Is the Benchmark

A 30 percent aided recognition score is strong in a category where the leader has 40 percent and weak in one where three competitors exceed 70 percent. The competitor set supplies the context that a published norm can’t.

That makes the brand list a methodological decision, not a formality. Leave out the category leader and every other figure loses its reference point.

The Gap Between Aided and Unaided Is Informative

A brand with high aided recognition and low unaided recall is known but not salient. People recognize it when prompted and don’t retrieve it when the need arises.

That pattern is typically more actionable than either number alone, because it points to salience instead of reach as the constraint. Compare your aided-to-unaided ratio with competitors' ratios in the same sample.

Treat Vendor Norm Tables With Caution

Some survey vendors publish awareness norms by industry. Before trusting one, look for a disclosed sample, field dates and a category definition.

A norm that can’t tell you who was asked, when, and about which category can’t tell you whether your number is good. Use it for curiosity, not for a target. The same caution applies to awareness figures quoted in competitors' marketing, which rarely state the population, the question or the date.

What Good Looks Like in Practice

A good single-read result is a clear ranking among the brands buyers actually consider, with intervals narrow enough that noise doesn’t reshuffle the ranking, and a foil rate low enough that aided recognition is credible.

A good result also answers the decision it was commissioned for. If the question was whether to lead a new market with brand spend, a clear answer to that question is the success criterion, whatever the absolute number.

Interpreting and Acting on the Result

Interpret awareness as a position relative to competitors among a defined population, then act on the pattern across measures instead of on any single figure. Three patterns cover most results.

Pattern 1: Low Aided, Low Unaided

The brand isn’t known. The constraint is generally reach, and the typical responses are distribution, paid media and partnerships that put the name in front of category buyers.

Before acting, check that the population was right. A B2B brand with low awareness among the general public may be well known among the 3,000 people who buy the category.

Pattern 2: High Aided, Low Unaided

The brand is recognized but not recalled. The constraint is salience, a brand’s propensity to come to mind in buying situations instead of simply being known.

Responses commonly include building links between the brand and specific buying situations, and consistent distinctive assets. A category entry point study is the natural next instrument.

Pattern 3: High Awareness, Low Consideration

The brand is known and not considered. The constraint is perception or fit, and more reach generally won’t help.

The next instrument is a brand perception study or depth interviews with aware non-considerers. Spending more on awareness here tends to buy recognition of a brand people have already decided against.

Segment Differences

Where segments were sized for, compare awareness across them with intervals. A brand strong in one region and weak in another has a geographic reach problem, which is often cheaper to fix than a category-wide one.

Cross-tabbing awareness against segments follows the same mechanics as any cross-tab analysis, with the paired-test caveat for brand-versus-brand comparisons within a segment.

When the Result Contradicts Expectations

A result that surprises the team deserves a methodological check before a strategic response. Confirm the population, the screener pass rate, the foil rate and the code frame, in that order.

If all four hold, the surprise is probably real. Teams commonly discover that their brand is less known than internal conversation suggested, because the people inside a company hear the brand name every day and buyers don’t.

What to Hand the Decision-Maker

Hand over one table: each brand’s unaided, top-of-mind and aided figures with intervals, the foil rate, the sensitivity-adjusted aided figure, and consideration among the aware. Add one sentence naming the pattern and the constraint it implies.

Then add the scope sentence, in the same document. The figures describe one population, in one market, at one time, measured on one brand list.

Running the Analysis With Claude or ChatGPT

A brand awareness analysis has two steps a language model can do well when the instructions are specific: coding unaided verbatims against a brand code frame, and computing the awareness table with intervals and paired comparisons. This chapter covers both, with copy-paste prompts.

General theming mechanics, the plain-versus-engineered prompt comparison and cross-tabbing across segments are covered in the launched guide to coding open-ended responses and the segment cross-tab guide. For the wider case for AI-assisted analysis and its limits, see how to analyze survey data with AI without sacrificing rigor. What follows is only what’s specific to awareness.

What the Analysis Produces

Two named artifacts. The first is a coded unaided file, with each respondent’s mentions mapped to brand codes in the order given, plus an uncoded bucket and a count of each.

The second is the awareness table: every brand’s unaided, top-of-mind and aided figures with Wilson intervals, the foil rate, the foil-excluded aided figure, and paired comparisons between your brand and each competitor.

Setup and Getting Your Data in

Export the responses as CSV and check a sample file to confirm how rows map to respondents, since the export is organized by response. The file needs a respondent identifier, the unaided open-text column, one column per aided brand coded 1 or 0, the “None of these” column, the two foil columns, and any segment fields.

Build the brand code frame before prompting. List each brand’s canonical name followed by every variant you expect: abbreviations, common misspellings, product names and former names.

Unaided brand coding is a deductive task, applying a known list instead of discovering themes. That distinction matters for accuracy, which the pitfalls section covers.

The default threshold of 50 in both prompts is a placeholder, not a published floor. At 50 respondents the 95 percent margin at 50 percent awareness is about 14 points, and 96 respondents give about 10, so set the threshold from the precision table in the sample size chapter.

Fuzzy matching on short brand names produces false matches at higher edit distances, so the coding prompt defaults to a cutoff of 1.

Prompt One: Coding Unaided Brand Mentions

What this prompt does: codes each respondent’s unaided brand mentions against a
fixed brand code frame, preserving mention order.
What it returns: a CSV named "unaided_coded.csv" and a summary table named
"Unaided coding summary".

The attached file has one row per respondent. The unaided answer is in column
[UNAIDED COLUMN, default "unaided"]. The respondent ID is in column
[ID COLUMN, default "respondent_id"]. The brand code frame is:
[BRAND CODE FRAME, default one line per brand written as "Canonical name: variant, variant, variant"]

Use code to calculate this, not estimation.

Rules:
1. Treat blank, whitespace-only and "none"/"n/a"/"don’t know" answers as
   non-responses. Keep those respondents in the file with no codes and count
   them separately. Don’t drop them from the denominator.
2. Split each answer into separate mentions in the order typed. Map each
   mention to a canonical brand only if, after lowercasing and trimming spaces
   and punctuation, it exactly matches a variant in the code frame. Put everything else in "UNCODED" with
   the original text preserved.
3. Top-of-mind is the first coded mention for each respondent. A respondent
   whose first mention is UNCODED has their first coded mention used instead,
   and you report how many respondents this applied to.
4. Any segment in [SEGMENT COLUMN, default "none"] with fewer than
   [THRESHOLD, default 50] respondents, including any segment with zero
   respondents, is reported as "insufficient respondents" rather than as a
   percentage.
5. Recount every brand’s mentions a second time using a different method: a
   case-insensitive fuzzy string match with an edit-distance cutoff of
   [EDIT DISTANCE, default 1] against the code frame variants, run as
   separate code. Compare the two counts brand by brand.
6. If the two counts differ for any brand, list the respondent IDs and raw
   text behind the difference. Don’t silently choose one count.
7. Output "unaided_coded.csv" with columns respondent_id, mention_order,
   raw_text, brand_code, and the "Unaided coding summary" table with each
   brand’s mention count, top-of-mind count, and the UNCODED count.

Prompt Two: The Awareness Table With Paired Comparisons

What this prompt does: computes unaided, top-of-mind and aided awareness with
intervals, the foil sensitivity check, and paired brand comparisons.
What it returns: one table named "Awareness table" and one named "Paired
comparisons".

Use the attached coded unaided file and the original response file. The coded file has one row per mention, so first collapse it to one row per respondent: a 1 or 0 flag per brand for any coded mention, plus the brand of the first coded mention. Then join it to the response file on
[ID COLUMN, default "respondent_id"]. Aided columns are [AIDED COLUMNS,
default all columns starting "aided_"], coded 1 if selected and 0 if not.
Foil columns are [FOIL COLUMNS, default "aided_foil1" and "aided_foil2"].
The focal brand is [FOCAL BRAND, default the first real brand column].

Use code to calculate this, not estimation.

Rules:
1. The denominator for every figure is all qualified respondents, including
   those who selected "None of these" or gave no unaided answer. Report the
   denominator. Exclude only respondents flagged in [EXCLUSION COLUMN, default
   "none"], and report how many. If a respondent left every aided column blank, treat them as an aided non-response, keep them out of aided figures only, and report the count.
2. Any figure based on fewer than [THRESHOLD, default 50] respondents,
   including zero, is reported as "insufficient respondents".
3. For each real brand, compute unaided, top-of-mind and aided shares with a
   95 percent Wilson score interval. Name the method.
4. Compute the foil rate: the share selecting at least one foil. Recompute
   aided shares excluding those respondents and show both.
5. For the focal brand against each competitor, compute the aided difference
   and a 95 percent interval using the paired formula: variance equals
   (share aware of focal only + share aware of competitor only minus the
   squared difference) divided by n.
6. Recompute every interval a second time with a different method: a
   bootstrap of [RESAMPLES, default 2000] resamples over respondents.
7. If the two methods disagree by more than [TOLERANCE, default 1]
   percentage point on any bound, report the disagreement and both values.
   Don’t reconcile it silently.
8. Output "Awareness table" and "Paired comparisons" as markdown tables.

Reading the Output

Read the coding summary before the awareness table, since coding errors frequently propagate unnoticed. A large UNCODED bucket means the code frame is missing variants, and every unaided figure downstream inherits the gap.

Read the UNCODED text itself. Real brands you didn’t list often appear there, and a competitor you left off the aided list is a design finding in its own right.

Then read the paired comparisons instead of eyeballing the separate intervals. Two brands whose individual intervals overlap can still differ significantly on the paired test, which the worked example in the analysis chapter shows.

What to Verify Before Reporting

Four checks, each quick:

Confirm the denominator matches the qualified respondent count in the raw file
Confirm the fuzzy recount and the model’s coding agree, or that disagreements are explained
Confirm the Wilson and bootstrap intervals agree within the stated tolerance
Confirm top-of-mind shares sum to no more than the share of respondents with any coded mention

The last check catches the most common coding error, which is a respondent counted as first-mentioning two brands.

Pitfalls

Six failure modes are documented for this kind of work, and each changes how the prompts above should be used.

Discovering themes is harder than applying them. Hill and colleagues, in PLOS Digital Health in April 2026, found models matched human analysts on deductive coding against an existing codebook, at 93.5 percent against 92.7, and were materially worse at inductive discovery. Brand coding against a fixed frame is on the favorable side of that finding, so prompt one supplies the frame instead of asking the model to find brands.

Rare codes get over-predicted. Ashwin, Chhabra and Rao, in Sociological Methods and Research in May 2025, found non-random bias in 10 of 19 codes, with sparse codes systematically over-predicted. For awareness, a small brand is a sparse code, so its unaided count deserves the closest check.

Outputs aren’t deterministic. Thinking Machines Lab reported in September 2025 that 1,000 completions at temperature zero produced 80 unique outputs. Neither consumer chat client exposes a seed, so any figure you intend to report should be produced twice.

Long inputs lose the middle. Liu and colleagues, in Transactions of the Association for Computational Linguistics in 2024, found models use information at the start and end of long contexts better than information in the middle. Paste the data first and the instructions and code frame after it, and split very large files.

Models tend to agree with the user. Sharma and colleagues, at ICLR 2024, documented sycophancy across assistants, and OpenAI withdrew a model update in April 2025 for being overly agreeable. Never tell the model which brand you expect to lead.

Verbatims can contain personal information. Unaided answers occasionally include names, employers or contact details. Check which data-use terms apply to the tier you’re using before uploading response files.

In Sprig: Claude or ChatGPT can reach studies and responses through the Sprig MCP connector, with access following each user’s account and workspace role. The limits: the documentation caps response retrieval at 1,000 responses at a time, and both prompts’ calculations run in the AI client, not a Sprig agent, so the verification checks above still apply.

Researchers remain responsible for checking the coded file and the intervals before any figure is reported.

The Critique You Should Know

The main critique of brand awareness measurement is that the standard instruments are biased against exactly the brands and buyers that growth depends on, and that aided recognition is inflated by people claiming to know brands they don’t. Both are supported by peer-reviewed evidence, and both have limits.

Unprompted Measures Skew Toward Brand Users

Romaniuk and Wight (2009), “The Influences of Brand Usage on Response to Advertising Awareness Measures,” International Journal of Market Research, found brand users about 2.5 times as likely as non-users to recall advertising on unprompted measures, compared with about 1.7 times when prompted with the brand.

They conclude that scores from different awareness measures shouldn’t be compared unless users and non-users are analyzed separately, and recommend brand-prompted measures for small or new brands.

For those brands, most responses under a brand prompt come from non-users. The study measured advertising recall, not brand recall, so the transfer to a brand awareness read is an inference. Romaniuk has also argued, in work the brand tracking guide cites, that top-of-mind measures bias against non-users and small brands, which points the same way.

If the skew holds, an unaided-only study of a small brand leans toward its existing customers' memory and reports it as market awareness.

You Can't Convert One Awareness Measure Into Another for a Single Brand

Romaniuk, Sharp, Paech and Driesener (2004), “Brand and Advertising Awareness: A Replication and Extension of a Known Empirical Generalisation,” Australasian Marketing Journal 12(3), 70 to 80, replicated the finding that top-of-mind, spontaneous and aided awareness are strongly related at category level. They also found that modelling a single brand over time was less successful.

So an unaided figure can’t be estimated from an aided one for your brand. If you need both, field both.

Aided Recognition Is Inflated by Over-Claiming

Paulhus, Harms, Bruce and Lysy (2003), “The Over-Claiming Technique: Measuring Self-Enhancement Independent of Ability,” Journal of Personality and Social Psychology 84(4), 890 to 904, established that claimed familiarity with non-existent items measures self-enhancement independently of real knowledge.

Brands are items like any other. A respondent who doesn’t want to appear uninformed, or who isn’t reading carefully, will recognize brands they’ve never encountered. The practitioner evidence in the quality control chapter points the same way. The critique isn’t that aided recognition is meaningless, but that it carries an unknown inflation unless foils are fielded to estimate it.

The Counter-Critique: Awareness Drives Choice in Low-Involvement Categories

A broader critique, common in practitioner writing, holds that awareness is a vanity metric with no link to choice. Two controlled experiments answer that critique directly, within limits.

Hoyer and Brown (1990), “Effects of Brand Awareness on Choice for a Common, Repeat-Purchase Product,” Journal of Consumer Research 17(2), 141 to 148, found brand awareness was a dominant choice heuristic among subjects who knew one brand in the choice set, even when other brands were of higher quality.

Macdonald and Sharp (2000), Journal of Business Research 48(1), 5 to 15, replicated the study with a different product and 462 participants. Consumers facing an awareness differential showed an overwhelming preference for the high-awareness brand despite quality and price differences, and decided faster while sampling fewer options.

Both were controlled studies of low-involvement, repeat-purchase grocery products. They support the claim that awareness matters for choice in categories like that, and they don’t license the same claim for a considered B2B purchase with a buying committee. A B2B team should read its awareness figure as the size of the pool it can compete for, not as a forecast of wins.

What the Critique Doesn't Say

None of this evidence says awareness surveys shouldn’t be run. It says the design choices covered earlier aren’t matters of taste.

Surveying category buyers answers the first critique, fielding both measures answers the second, and foil brands answer the third. A survey built that way has already absorbed the critique. The counter-critique matters most for consumer brands: in a repeat-purchase category, a measured awareness gap is a plausible reason for a share gap.

Common Mistakes

Eight mistakes recur in brand awareness surveys, in rough order of how much damage they do. Each has a tell, the thing to look for when you suspect it.

Surveying Your Own Users and Calling It Market Awareness

The most common and most consequential. An email to your customer list returns awareness near 100 percent, and a board slide reports it as the brand’s market position.

The tell is an aided recognition figure far above any competitor’s. Check the sample source before believing it.

Showing the Brand List Before the Recall Question

Any brand name shown before the unaided question primes recall, and the damage can’t be undone at analysis. This includes a brand name in the survey title, the invitation email or the screener.

The tell is an unaided figure suspiciously close to the aided one for your own brand. Read the invitation copy and screener before fielding.

Comparing Two Brands With a Two-Sample Test

Brand scores from the same respondents are paired. The two-sample test typically overstates the uncertainty of the gap and hides real leads, as the worked example showed.

The tell is a report stating that two brands are “statistically tied” with no mention of which test was used.

Leaving Out Foil Brands

Without foils, the inflation in aided recognition is unmeasured. Every aided figure carries an unknown amount of over-claiming, and there’s no way to estimate it afterwards.

The tell is an aided list containing only real brands. Two invented names cost almost nothing to add.

Coding Unaided Answers Carelessly

Misspellings, product names and abbreviations that aren’t mapped to their brand reduce its unaided count silently. The error rarely shows in the final table.

The tell is a large uncoded bucket, or a brand whose unaided figure looks implausibly low against its aided figure. Read the uncoded text.

Excluding Non-Answers From the Denominator

Dropping respondents who typed nothing or selected “None of these” inflates every brand’s awareness. It feels like cleaning the data and it’s changing the population.

The tell is a denominator that doesn’t match the qualified respondent count.

Reporting Segments the Sample Wasn’t Sized for

A 500-complete study reported across five regions commonly produces five figures with margins near 10 points. Differences between them are frequently noise presented as insight.

The tell is a segment table without intervals or counts. Add both, and most of the apparent differences dissolve.

Changing the Brand List During Fieldwork

Adding a competitor halfway through fieldwork because a stakeholder asked for it splits the sample into two instruments. Aided figures for every brand then mix respondents who saw different lists.

The tell is a brand with fewer aided responses than the rest. Freeze the list at soft launch and add requests to the next study.

Treating a Single Read as a Trend

Comparing this year’s survey with last year’s, on a different list or panel, and reporting the difference as growth. Different instruments measure different things, even when the question sounds the same.

The tell is any before-and-after chart where nobody can confirm the two studies used identical wording, list, population and mode.

Synthetic Respondents

Synthetic respondents can’t produce a meaningful brand awareness figure. Awareness is a memory measure, and a language model has no memory of your category to recall or fail to recall.

A model asked which brands come to mind returns brands that are prominent in its training data. That’s a measure of a brand’s presence in public text, not in the heads of category buyers in a specific market.

The general evidence on synthetic survey data reinforces the point. Bisbee and colleagues (2024), in Political Analysis, compared synthetic responses against a large real sample and found averages could look close while regression coefficients frequently differed, with variance artificially deflated. This guide doesn’t recommend synthetic respondents for any awareness measure.

What a Synthetic Run Can Check

A synthetic pass over the draft questionnaire can surface confusing screener wording, a category cue that invites off-category answers, and an aided question that reads ambiguously. Those are properties of the instrument, not of a population.

Treat anything a synthetic run returns about brands themselves as noise. The only awareness figures worth reporting come from people who were asked.

Alternatives and Adjacent Methods

Several instruments answer questions a brand awareness survey can’t. Choose by the question, not by the phrase the request arrived in.

Brand Tracking

Use a brand tracker when the question is whether awareness is changing. It uses the same core items with a frozen instrument, constant sampling and wave-over-wave tests.

Brand Lift Studies

Use an exposed-versus-control lift study when the question is whether a campaign raised awareness. It supplies the counterfactual a single read lacks.

Brand Perception and Category Entry Points

Use a brand perception study when the question is what people think of you, and a category entry point battery when the question is which buying situations bring you to mind.

Usage and Attitudes and Competitive Analysis

Use a usage and attitudes study when awareness is one input among several, alongside category habits, satisfaction and switching intent.

Message Testing and MaxDiff

Use message testing when the question is which claim or creative to run, and MaxDiff when the question is which of many claims matter most.

Depth Interviews

Use qualitative depth interviews when the question is why buyers don’t know or don’t consider you. Interviews are a qualitative instrument, not a survey tool, and they answer the reasoning question awareness can’t reach.

Choosing Between Them

Most teams need two of these instruments, not one. An awareness read paired with a perception study explains both how many buyers know the brand and what the aware ones think of it, which covers the three interpretation patterns in one fieldwork cycle.

Where budget allows only one study, choose the instrument that answers the decision already on the table, and write the others into the plan for later.

Frequently Asked Questions

What Should a Brand Awareness Survey Ask?

A brand awareness survey should ask, in order: a category screener, an unaided recall question in open text, an aided recognition question showing your brand and competitors plus two fictitious foil brands, and optionally consideration among recognized brands. Unaided must come before aided, because any brand list shown first contaminates recall.

The instrument spec in this guide lists every item in fielding order, and Sprig’s Design Agent can build it as a study.

How Many People Do You Need for a Brand Awareness Survey?

No methodology body publishes a minimum sample for brand awareness surveys.

Size from the margin of error your decision needs: at 50 percent awareness, 385 qualified completes give about plus or minus 5 points at 95 percent confidence, and 1,068 give about plus or minus 3. Each reported segment needs that sample if it must be read at the same precision.

Should I Survey My Customers or the Whole Category?

Survey category buyers for any market-level awareness figure. Your own customers already know you, so their awareness is close to universal and says little about the market. Romaniuk and Wight (2009) found brand users about 2.5 times as likely as non-users to recall a brand’s advertising on unprompted measures, which suggests a similar skew in brand recall. Survey customers only when awareness among customers is the actual question.

How Is a Brand Awareness Survey Different From a Brand Tracker?

A brand awareness survey measures the level of awareness once. A brand tracker measures change over time by repeating the same frozen instrument, sample definition and mode at intervals.

If you might measure again, build the awareness survey as wave 1 of a tracker from the start, following the brand tracking guide.

How Do You Measure Brand Awareness Before and After a Campaign?

A before-and-after awareness survey can’t isolate a campaign’s effect, because anything else that happened in between also moves awareness. Measure campaign impact with an exposed-versus-control design, where people who saw the campaign and comparable people who didn’t are surveyed at the same time. Advertising platforms call this a brand lift study.

What Is Top-of-Mind Awareness, and Should I Report It?

Top-of-mind awareness is the share of respondents who name your brand first when asked, unprompted, which brands come to mind in a category. Report it for established brands. For small brands it’s often unstable, because a few first mentions move it several points, and like all unprompted measures, it leans toward existing users.

Why Include Fake Brands in a Brand Awareness Survey?

Fictitious foil brands measure over-claiming. Any respondent who selects a brand that doesn’t exist is recognizing something they can’t have encountered, which reveals how much aided recognition is inflated. Report the share selecting any foil, and show aided figures with and without those respondents.

How Do I Compare My Brand's Awareness With a Competitor's?

Use a paired comparison, because both scores come from the same respondents. The variance of the difference depends on respondents who know one brand but not the other.

A two-sample test usually overstates the uncertainty and can hide a real lead: in this guide’s worked example, a four-point gap was significant under the paired test and not under the two-sample test.

Can I Run a Brand Awareness Survey in B2B?

Yes, a B2B brand awareness survey works when the population is defined by role and decision involvement instead of job title alone.

Screen for people who influenced or made a recent purchase decision in the category. B2B samples are often smaller and generally costlier, so decide segment cuts before sizing.

Can AI Code Unaided Brand Recall Answers?

Yes, a language model can code unaided brand mentions well when given a fixed brand code frame, because that’s deductive coding, where one peer-reviewed comparison found models roughly matched human analysts.

Supply the frame, require an uncoded bucket, and recount with a separate fuzzy-match method. Researchers remain responsible for reading the uncoded bucket and the recount differences before any figure is reported.

How Long Does a Brand Awareness Survey Take to Run?

A brand awareness survey typically takes a few weeks from study plan to report, and fieldwork itself is often a matter of days. Category incidence drives the field time more than anything else: a category bought by most adults fills quickly, while a niche B2B role can take much longer and cost more per qualified complete. Get a feasibility estimate for your screener before committing to dates.

Can I Run a Brand Awareness Survey Across Several Countries?

Yes, a multi-country brand awareness survey works when each market is treated as its own population with its own brand list, category cue in local buyer language, and locally plausible foil brands. Translate the category cue with care, since a literal translation often describes a different category. Compare markets only on brands fielded in both, and size each market separately.