What it's for, and what it isn't
Use it when you need rates for one campaign and one message: how many people recognised it over and above the false-claim floor (the share of people who say they've seen an ad they couldn't have seen), how many knew it was yours, how many took out the message you meant, and what they say they did next. It helps you adjust media, placement or creative with numbers you can export.
It isn't a brand lift, incrementality or ROI study, and it doesn't re-measure awareness or consideration (your Tracksuit tracker does that). It doesn't benchmark your ad against norms or choose between ads that haven't run yet. If you need to know why the campaign landed, use Campaign diagnostic.
The study type it matches
A post-campaign ad recognition study, often called a post-test or ad tracking study, with a false-claim control. Measuring recognition against people who couldn't have seen the ad goes back to the 1940s (Lucas, 1942). It reports the classic chain of advertising response: noticed, linked to the brand, understood, acted on (Lavidge & Steiner, 1961). This is the order we report in, not a theory of how ads work (Vakratsas & Ambler, 1999).
How it runs
- Who: people in the market the campaign was meant to reach (for example, planning to buy in your category in the next 12 months). We never screen on whether they remember the ad, because that's what we're measuring. We screen out people who work in marketing, advertising or research, and record whether each person owns or uses your brand now, used to, or never has.
- Sample: about 300 people in markets where the campaign ran. Everyone sees two foil ads: real ads that didn't run in this market. About 1 in 8 people from the campaign markets (about 38), chosen from those who missed the ad, named the wrong brand or were confused, take part in a short follow-up interview.
- Mode: a short online questionnaire. Follow-ups are interviews moderated by AI, typed, or spoken where available.
- Timing: in the final week of the campaign or within about two weeks of it ending. The "seen in the last few weeks" period covers the whole campaign plus the gap to fieldwork, is the same for everyone, and we record days since the campaign ended.
- Question order:
- Which ads they remember from the last few weeks, and what happened in them, with nothing shown
- Brand-masked stills of your ad (logo and end card removed), mixed in random order with foil ads (real ads that didn't run in this market): seen, not seen or not sure, then where
- Whose ad it was, asked on the masked stills, first unprompted and then from a short list
- What the ad was telling them, in their own words
- What they've done since seeing it, including "none of these"
- Short follow-up interviews with some people who missed the ad, named the wrong brand or were confused
Questions 3 to 5 are asked of people who recognised the ad.
Why it's designed this way
- Foils in every study. Some people always claim to recognise ads they couldn't have seen. Measuring against ads they couldn't have seen is the established correction (Lucas, 1942; Appel & Blum, 1961; Singh & Churchill, 1986).
- Real recognition tracks reach. Recognition of a campaign's real creative rises with the media weight behind it (Southwell et al., 2002).
- Memory before prompting. We ask unprompted recall before showing anything, because earlier questions shape later answers (Schuman, Presser & Ludwig, 1981), and recall and recognition measure different things (du Plessis, 1994).
- Logos masked. With the logo visible, "whose ad was it?" tests reading, not memory.
- Owner status recorded. A brand's users remember its ads more than non-users on every measure tested, so a campaign aimed at new buyers should be judged on non-users (Vaughan, Beal & Romaniuk, 2016).
- Baselines are a floor, not a lift measure. Observational comparisons often misstate ad effects even with extensive controls (Gordon et al., 2019). Even randomised tests need very large samples to measure returns precisely (Lewis & Rao, 2015). Credible lift needs a randomised experiment, such as a platform brand lift study or a randomised geo test (Vaver & Koehler, 2011). That's why we use foils only to correct recognition, and why we never claim lift.
- Short questionnaire, depth in the follow-ups. Long questionnaires wear down answer quality (Krosnick, 1991).
- A codebook written before fieldwork. Open answers are coded to fixed categories, and agreement between coders is measured with Krippendorff's alpha (Hayes & Krippendorff, 2007).
- Field while it's fresh. A short, intense burst is forgotten quickly once it stops (Zielske, 1959; Simon, 1979).
How to read the results
- Three levels from the same run. An Exec brief (a short read on recognition, brand linkage, message takeout and claimed action, with a next step), a Stakeholder report (every finding with its evidence, plus quotes on the misses) and a Deep dive (everything, with bases and the method note).
- Recognition is always net: your ad minus the foils. We never show recognition on its own.
- Brand, message and action results are among people in campaign markets who recognised the ad.
- Bases: with 100 or more recognisers we show percentages. With 50 to 99 we show directional results with their precision. Below 50, answers to the fixed questions are shown as "12 of 40", always with the base, and marked directional.
- Owners and non-owners are shown separately when each group has at least 50 recognisers. Otherwise the owner share sits next to every rate.
- Actions are what people say they did. They're labelled "say they…", they aren't verified, and we never compare them between groups.
- Precision is modelled precision. It assumes simple random sampling, which an online panel isn't, so it's a guide and not a margin of error (AAPOR). As a guide, a 50% result carries about ±10 points at 100 recognisers and about ±14 points at 50.
- Follow-up interviews explain the misses. They're quotes and themes, and a theme that comes only from follow-up questions is never counted.
- What it doesn't claim: lift, that the campaign caused anything, sales or ROI, a change in awareness or consideration, or a score against norms. Every number it reports is one the design can stand behind.
Quality
- Everyone gets exactly the same questionnaire.
- Open answers are coded to the pre-written codebook by two coders, and Krippendorff's alpha is reported before any percentage reaches you. AI-assisted coding is used only alongside a random sample checked by a person.
- A named Tracksuit researcher reviews and releases every report. The method note lists the groups, dates, how many people qualified, codebook and alpha.
- Respondents come from established online research panels and are screened before they take part.
- How we use AI. The main questionnaire is a fixed form. In follow-up interviews, the AI interviewer asks the fixed questions in order and may add one or two neutral follow-ups per question. It's set up not to skip required questions or bring in new topics. People design the study, approve the questionnaire and interview guide, check the coding and release the report. Rates are calculated from the answers, and the written summary is set up not to add a finding the answers don't support. Respondents consent before they start, can stop at any time and can ask for their data to be deleted. We don't use synthetic respondents.
Where Tracksuit's own judgement sits
These are Tracksuit's standards. They're set from public panel-cost norms, standard sampling arithmetic and the evidence above, and we check each one in a small test launch before full fieldwork.
- Market floor: we run the study when at least 10% of people starting the survey fit the target market. Between 5% and 10% we scope it with you. Below 5% we don't run it.
- Reporting bases: percentages at 100 or more recognisers, directional results at 50 to 99, and below 50, "12 of 40" with the base, marked directional.
- Sample design: about 300 people in campaign markets, two foil ads for everyone, an optional control of about 150 people in markets the campaign didn't reach (extra cost), and about 1 in 8 in follow-up interviews.
- Timing: within about two weeks of a burst campaign ending, and four weeks at most.
- Campaign fit: we qualify campaigns on reach (channels, markets, dates, GRPs or impressions), not budget. A single-platform digital ad is best measured by that platform's own brand lift study.
References
- American Association for Public Opinion Research (AAPOR). Understanding a "credibility interval" and how it differs from the "margin of sampling error" in a public opinion poll. https://aapor.org/statements/understanding-a-credibility-interval-and-how-it-differs-from-the-margin-of-sampling-error-in-a-public-opinion-poll/ (opens in a new tab)
- Appel, V., & Blum, M. L. (1961). Ad recognition and respondent set. Journal of Advertising Research, 1(4), 13–21. https://doi.org/10.1080/00218499.1961.12519628 (opens in a new tab)
- du Plessis, E. (1994). Recognition versus recall. Journal of Advertising Research, 34(3), 75–91. https://doi.org/10.1080/00218499.1994.12466952 (opens in a new tab)
- Gordon, B. R., Zettelmeyer, F., Bhargava, N., & Chapsky, D. (2019). A comparison of approaches to advertising measurement: Evidence from big field experiments at Facebook. Marketing Science, 38(2), 193–225. https://doi.org/10.1287/mksc.2018.1135 (opens in a new tab)
- Hayes, A. F., & Krippendorff, K. (2007). Answering the call for a standard reliability measure for coding data. Communication Methods and Measures, 1(1), 77–89. https://doi.org/10.1080/19312450709336664 (opens in a new tab)
- Krosnick, J. A. (1991). Response strategies for coping with the cognitive demands of attitude measures in surveys. Applied Cognitive Psychology, 5(3), 213–236. https://doi.org/10.1002/acp.2350050305 (opens in a new tab)
- Lavidge, R. J., & Steiner, G. A. (1961). A model for predictive measurements of advertising effectiveness. Journal of Marketing, 25(6), 59–62. https://doi.org/10.1177/002224296102500611 (opens in a new tab)
- Lewis, R. A., & Rao, J. M. (2015). The unfavorable economics of measuring the returns to advertising. The Quarterly Journal of Economics, 130(4), 1941–1973. https://doi.org/10.1093/qje/qjv023 (opens in a new tab)
- Lucas, D. B. (1942). A controlled recognition technique for measuring magazine advertising audiences. Journal of Marketing, 6(4, part 2), 133–136. https://doi.org/10.1177/002224294200600431.1 (opens in a new tab)
- Schuman, H., Presser, S., & Ludwig, J. (1981). Context effects on survey responses to questions about abortion. Public Opinion Quarterly, 45(2), 216–223. https://doi.org/10.1086/268652 (opens in a new tab)
- Simon, J. L. (1979). What do Zielske's real data really show about pulsing? Journal of Marketing Research, 16(3), 415–420. https://doi.org/10.1177/002224377901600315 (opens in a new tab)
- Singh, S. N., & Churchill, G. A. (1986). Using the theory of signal detection to improve ad recognition testing. Journal of Marketing Research, 23(4), 327–336. https://doi.org/10.1177/002224378602300402 (opens in a new tab)
- Southwell, B. G., Barmada, C. H., Hornik, R. C., & Maklan, D. M. (2002). Can we measure encoded exposure? Validation evidence from a national campaign. Journal of Health Communication, 7(5), 445–453. https://doi.org/10.1080/10810730290001800 (opens in a new tab)
- Vakratsas, D., & Ambler, T. (1999). How advertising works: What do we really know? Journal of Marketing, 63(1), 26–43. https://doi.org/10.1177/002224299906300103 (opens in a new tab)
- Vaughan, K., Beal, V., & Romaniuk, J. (2016). Can brand users really remember advertising more than nonusers? Testing an empirical generalization across six advertising awareness measures. Journal of Advertising Research, 56(3), 311–320. https://doi.org/10.2501/JAR-2016-037 (opens in a new tab)
- Vaver, J., & Koehler, J. (2011). Measuring ad effectiveness using geo experiments. Google Research. https://research.google/pubs/measuring-ad-effectiveness-using-geo-experiments/ (opens in a new tab)
- Zielske, H. A. (1959). The remembering and forgetting of advertising. Journal of Marketing, 23(3), 239–243. https://doi.org/10.1177/002224295902300301 (opens in a new tab)