An ad can stay with someone because of humor, useful information, storytelling, music, or another distinctive detail. Which device stands out depends on the audience and the creative. A viewer may remember a joke or character but connect it to the wrong advertiser, so memorability alone is not enough.
The practical answer is to measure several outcomes separately: whether people noticed the ad, what they recall without a prompt, whether they identify the correct brand, and whether they understand the main message. Then use separate research to assess consideration, trust, behavior, or revenue.
Available survey evidence can suggest creative hypotheses, but it cannot establish that humor, education, production quality, or any other device caused recall or business impact. The workflow below turns that distinction into a usable research design.
What makes an ad memorable?
In a HubSpot survey of 281 U.S. consumers, respondents selected one primary reason they remembered a recent advertisement. Humor was the most frequently selected reason, followed by the value of the information presented. Those findings are useful starting points for creative testing, not proof that humor or education causes recall, sales, or long-term memory.
HubSpot identifies the underlying research as 2021 survey data, while the article page was updated on August 17, 2026. Keep the research date separate from the page update date when citing the finding. The accessible article does not clearly state the percentage for the second-ranked information response, so do not infer or publish one from the surrounding categories.
What the memorability research actually tells you
The HubSpot respondents selected one reason, which means each figure describes the primary reason chosen in this survey. It does not measure the independent effect of a creative device, and it does not capture secondary influences. HubSpot reports product value at 7%; an interesting story and a catchy element at 4% each; interesting characters and high action at 3% each; sadness, topic relevance, and a celebrity or expert at 2% each; other reasons at 8%; and perceived high or low budget at 1% each.
The result is directional. The accessible article does not provide full sampling, recruitment, weighting, field-date, questionnaire, or margin-of-error details. A small, single-answer survey cannot establish population-wide preferences or causation. In particular, few respondents selecting perceived budget as their primary reason does not show that production quality has no effect on credibility, attention, conversion, or long-term memory.
Clutch provides a separate and older point of comparison. Its 2018 survey of 1,030 U.S. consumers found that 53% considered humorous advertisements memorable and enjoyable. That is a reported consumer attitude, not an experiment showing that humor caused recall or sales. Clutch also reported that 51% disliked ads because they did not learn anything. That is a stated reason for disliking an ad, not a direct recall measure. See the Clutch survey and methodology.
Other examples answer different questions. HubSpot’s roundup of nine YouTube ads is an editorial selection, not a representative sample. Kellogg’s Super Bowl Advertising Review applies the ADPLAN framework, including Attention, Distinction, Positioning, Linkage, Amplification, and Net Equity. Its grades are strategic evaluations, not consumer-memory measurements. See the Kellogg results archive.
The General is also an example of why production-quality conclusions require care. The company said in a 2021 release that it was addressing consumer misperceptions associated with its advertising through a brand refresh. The release supports the company’s stated response, but it does not independently prove that low-budget production caused those perceptions or that the refresh corrected them.
A respondent-selected reason is a clue about what stood out, not proof of what caused recall or business impact.
Separate recall from recognition, brand linkage, and effectiveness
Use distinct measures because each answers a different question:
- Attention: Did the person notice the ad?
- Unaided recall: Can the person remember an ad without seeing it again?
- Recognition: Can the person identify an ad or brand when given a prompt?
- Brand linkage: Does the person connect the creative to the correct advertiser?
- Message comprehension: Can the person describe the main product, benefit, or claim?
- Brand and business outcomes: Did consideration, trust, behavior, or revenue change?
A viewer can remember a song, joke, or character but name the wrong brand or fail to identify the product. Record unaided recall before showing any logo, ad, or brand list. Then ask which brand the ad promoted and what its main message was. Report each result separately rather than combining them into an undefined memorability score.
Design a survey that produces interpretable ad-recall data
Keep the question order consistent and preserve the exact wording. A compact instrument can ask:
- “Which recent advertisement do you remember?” Record the unaided answer, including no-recall responses.
- “What do you remember about it?” Capture open text without showing the creative.
- “Which brand do you think it was for?” Record brand identification separately.
- “What was the main message?” Record the respondent’s description of the claim or benefit.
- “What is the primary reason it stayed with you?” If replicating a single-select format, require one answer from a defined list and retain the original wording.
For each response, retain the survey wave, field date, geography, ad or creative-variant ID, question ID and version, response ID, and whether the measure was unaided or prompted. Preserve raw “other” answers for later review. Version the category list because changing a category between waves can make results incomparable. Flag small subgroup counts, and do not treat sample size alone as evidence that a sample represents a wider population.
A practical workflow for coding and reporting responses
Use the approved survey dataset or research response store as the system of record. This is a proposed research workflow, not a documented HubSpot, CRM, or automation integration. Organizations documenting a repeatable research workflow should keep individual answers, coding notes, ad records, source citations, and calculated summaries as distinct records.
One response row should mean one respondent’s answer to one question in one survey wave. If recall, brand identification, and message comprehension are separate questions, store separate question-response records linked by an anonymized response grouping key. Do not use a campaign total as an individual observation, and do not write a survey result directly into a CRM contact or deal event without preserving the source response and research basis.
For example, a closed-list reason response could be represented as follows. This is an illustrative research record, not a vendor-published schema:
{
"response_id": "resp-1042",
"survey_wave_id": "wave-2026-10",
"respondent_id_hash": "anon-7f3a",
"ad_id": "ad-variant-07",
"question_id": "memory-reason",
"question_version": "memory-reason-v1",
"respondent_region": "US",
"memorability_reason": "humor",
"response_timestamp": "2026-10-12T14:30:00Z"
}At this grain, a proposed uniqueness key is survey_wave_id + respondent_id_hash + question_id, subject to the research platform’s respondent model. Validate that the wave, ad, question version, timestamp, and selected category exist. Enforce one answer only when the instrument is single-select. Reject or separately classify a response with no recalled advertisement rather than assigning it a memorability reason.
When a respondent selects a fixed category, save that selection exactly as submitted. Use optional AI only to suggest a category for ambiguous free text, and never to reinterpret an explicit answer or infer campaign effectiveness.
If open-text “other” answers need coding, an analyst can use AI to suggest a category from an approved taxonomy and identify the words supporting it. Keep the raw response unchanged and store the suggestion in a separate annotation with the evidence span, confidence, model version, prompt version, and review status. A human analyst approves ambiguous, multi-reason, sensitive, or out-of-taxonomy answers. New categories should be documented as post hoc coding.
Calculate summaries deterministically from eligible, approved response records. An aggregate row should represent one survey wave, ad, audience or geography scope, question version, and response category. Store its count, eligible denominator, percentage, and calculation timestamp. Exclude duplicate response IDs and state the denominator. For concurrent writers, enforce a database unique key across the full aggregate grain and use a transactional upsert. A lookup followed by an insert is not race-safe.
Keep the aggregate separate from the raw observation. A proposed aggregate key is survey_wave_id + ad_id + respondent_region + question_version + memorability_reason; include additional audience dimensions whenever they define the reported scope. A citation record is separate again because one source may support several claims and should never substitute for a response ID.
Choose creative tests that fit the audience and objective
Humor and useful information are creative approaches to test, not universal prescriptions. Clutch’s finding describes what respondents said about humorous ads; it does not show that humor caused sales. For a humorous concept, check whether the intended audience understands the joke, remembers the advertiser, and can still explain the product benefit. For an educational concept, test whether people recall the information and accurately describe the product or claim.
- Can the intended audience understand the joke without extra context?
- Can viewers identify the brand and product after the creative?
- Does the product benefit survive the humor?
- Could a sensitive topic or vulnerable group become the punchline? If so, obtain human brand, legal, cultural, or inclusion review.
- Are exposure conditions and question order consistent across creative variants?
If a joke is remembered but brand linkage is weak, revise the brand cue or test the ending separately. If an educational ad is understood but not recalled, investigate the creative device rather than assuming more information will solve the problem. If variants are compared, define the audience, exposure conditions, field period, and measurement sequence before interpreting a difference.
How to interpret results without overclaiming
Before publishing a result, state what was measured, who was surveyed, where and when the survey ran, the exact question and answer rule, the denominator, and the source. Include the question version and whether the measure was unaided recall, aided recall, recognition, enjoyment, or perceived memorability. Label a respondent opinion as an opinion, an editorial selection as curated, and a strategic grade as an evaluation.
Say “few respondents selected perceived budget as their primary reason for remembering an ad in this survey,” not “budget does not matter.” Similarly, a low celebrity-selected response does not prove that celebrity use cannot affect attention, brand fit, or sales. These categories describe what respondents chose as their main reason in one small survey.
Use recall and brand linkage to diagnose creative performance. Use separate, fit-for-purpose research to assess preference, trust, behavior, or revenue. A memorable scene is useful only in relation to the campaign objective, the audience, and whether people connect it to the right brand and message.
Sources and reporting standard
Primary and comparative sources include HubSpot’s memorable-ads survey article, Clutch’s 2018 consumer advertising survey, Kellogg’s Super Bowl Advertising Review archive, and The General’s 2021 company release.
For every published statistic, preserve the source URL, publisher, author if available, publication or update date, survey field date, sample and geography, exact wording, response mode, category definitions, access date, and editorial verification status. That provenance makes later waves comparable and prevents a descriptive recall statistic from being mistaken for evidence of effectiveness.