RESEARCH PROPOSALRESEARCH PLAN

Can AI understand the product without weakening the campaign?

We propose keeping the campaign intact, changing supporting explanations, and measuring human and AI responses separately.

Document version
Draft 01
Updated
Design status
Design proposal / pre-experiment
Schedule
To be determined

What we want to learn

Research question and hypothesis

Without changing the mood of an advertisement, can better use cases and product explanations improve AI accuracy while preserving customer appeal?

Hypothesis

Material that connects the same product facts to usage situations may help AI explain the product accurately. Whether customer appeal is preserved must be checked separately.

This is an expected direction, not an established finding.

Why ask this question?

Advertising can create desire, while reviews and product descriptions help people decide whether a product fits. This test asks whether AI-oriented explanation requires changing the advertisement itself.

What we will compare

Keep the campaign
change the explanation

Keeps the advertisement unchanged while comparing different explanations.

Held constantSame advertisement · same product facts
A

Current explanation

Provide the advertisement and current product description as the baseline.

B

Situation-linked explanation

Keep the advertisement unchanged and connect the same facts to use cases.

Measure each condition separately
Check with peopleIs appeal preserved?

Measure understanding, appeal, and purchase consideration after customers read the material.

Check with AIDoes the explanation become more accurate?

Provide the material directly and compare product explanations and recommendation reasons for the same customer situation.

Human response and AI output are measured separately. One does not stand in for the other.

Controls

Keep product facts and the advertisement the same and limit the change to explanation structure. Record differences in length and reading time. If new facts are required, redesign to separate fact additions from structure effects.

Example: travel luggage

For a fictional suitcase, the ad shows a spontaneous weekend trip with friends. Supporting material explains what was packed, how it was carried, its size, and cautions. The actual brand and product are not yet selected.

How we will proceed

Proposed sequence

  1. Prepare comparison materials

    Select the product, customer situation, and advertisement, then version A/B materials and lock facts and evaluation criteria before viewing answers.

  2. Compare under matched conditions

    Consider random assignment for customers. For AI, mix order and repeat with the same question, model, and search setting.

  3. Review outcomes separately

    Where possible, blind the condition. Retain AI answers, customer responses, and errors, including results that do not improve.

This sequence is a draft. Sample size, repetitions, and decision rules will be fixed before execution.

What we will record

Measures and limits

MeasureMethodLimit
AI explanation accuracyCheck errors, omissions, and invented benefits against the product fact listAn accurate explanation does not imply customer preference
Recommendation relevanceCheck whether reasons and poor-fit conditions match the usage situationRepeating brand or ad copy is not success
Customer understanding and appealCollect open descriptions of what was remembered and who fits, then separately measure appeal and purchase considerationDo not substitute AI role-play for customer responses
Reading burdenMeasure reading time and explanations seen as difficult or unnecessarySeparate the effect of added length from structure

How we will judge results

Rules fixed in advance

AI explanation improves and customer appeal is maintained

Test whether the result repeats for other products and situations

AI explanation improves but customer appeal falls

Check whether length or expression created reading burden

Only customer response improves or both outcomes are unclear

Separate customer and AI effects and distinguish no effect from insufficient evidence

Numerical thresholds, acceptable decline, and chance-difference checks must be set before the experiment. An unclear difference is not evidence that appeal was maintained.

Decisions before launch

Items not yet fixed

Product and customer
Brand, product, and recruitment target not set
AI and questions
Service, model, and question wording not set
Scale and decision
Participants, repeats, and thresholds not set
Execution
Schedule, owner, budget, and consent process not set

Scope of this plan

  • The first proposal supplies materials directly; discovery and citation after web publication require a separate study.
  • This comparison cannot establish sales growth or market-wide change, and cost effectiveness and cross-cultural response remain unknown.
  • This is a design draft, separate from the monthly experiments already scheduled.

Background and sources

Reviewed guidance
and research rationale

Google guidance on AI features in Search

Background on standard search requirements, making important information available as text, and using images and video. It does not prove the proposed comparison will work.

Checked 2026-09-26

Read the full backgroundFull explanation and hypothetical example

A campaign can still create the desire to leave

Making a product easier for AI to explain does not require turning the campaign into a specification sheet. Keep the campaign’s appeal and supplement concrete information elsewhere for customers and AI to consult.

Imagine an ad for a travel bag. Friends leave for an unplanned weekend and walk through unfamiliar streets, making viewers feel, “I want to go too.” Adding dimensions and packing conditions to every scene could change the intended mood.

The campaign can express the pleasure of travel. A trip story can show what was packed. A product page can explain dimensions and cautions. This is the division of roles proposed here.

Google says that its AI search features do not require a special AI-only technique. It recommends useful content for people, established search fundamentals, important information in text, and appropriate images and video. It does not instruct advertisers to replace emotional expression with specification copy. Google guidance ↗

Google’s guidance cannot be assumed to apply identically to every AI service, and following it does not guarantee search visibility.

Give campaigns, stories, and product pages distinct jobs

An ad and a product page do not need the same voice. But the experience promised by the campaign should not conflict with the experience of using the product.

For a hypothetical luggage brand, the roles might be divided as follows:

MaterialRoleTravel-bag example
CampaignCreate the desire to leaveFriends taking an unplanned weekend trip
Use caseShow how the product is usedWhat someone packs and how they move on a short trip
Product pageExplain capability and cautionsCapacity, dimensions, care, and limitations
Customer reviewDescribe lived experienceThe trips where it was convenient and what was difficult

There is no need to create all four from scratch. Start with existing material that supports the campaign promise and fill only the missing explanation. Reviews must come from real customers; do not manufacture praise or hide inconvenient experiences to fit the campaign.

Do not reduce “free travel” to bag weight

For one person, free travel may mean light luggage. For another, it may mean leaving without a fixed itinerary or behaving differently from everyday life. Reducing all of this to “the bag is light” preserves only part of the intended story. A company cannot determine customer meaning internally.

Three questions make the relationship between an abstract campaign line and the product easier to examine:

  • What feeling should the campaign create? For example, “I want to leave lightly without a plan.”
  • Who needs it, and when? For example, someone who often takes short weekend trips.
  • What does the product actually help them do? Check whether it holds what the trip requires, is convenient to move with, and has any relevant drawbacks.

This is still our proposed explanation. Interviews or actual use are needed to establish whether frequent weekend travelers really value light packing and mobility. A plausible explanation is not evidence of customer understanding.

Find the missing piece before creating more content

Choose work based on the observed problem rather than increasing the volume of copy by default.

Observed problemCheck firstDo not rush into
Customers interpret the campaign differentlyWho saw which scene and languageProducing large amounts of AI-targeted copy without knowing the cause
Customers understand the ad but credible examples are missingActual use cases and product experienceMaking benefits larger without evidence
AI states product facts incorrectlySources used, product identity, and freshnessRebuilding the entire campaign
Customers understand and AI descriptions are accuratePreserve current material and test other situationsAdding material only because “AI requires it”

This table helps decide where to look first. The proposed response still needs evidence and testing.

Revise one campaign and test people and AI separately

Start with one campaign and one usage situation. Leave the ad intact and supplement either a travel use case or product explanation. Ask customers who see the existing and revised material how they understand the product and why they would choose it. Separately, give AI the same customer situation and observe whether its explanation and recommendation rationale change.

Two tests must remain distinct. Supplying material directly to an AI tests whether it can read and explain that material. It does not test whether an AI will discover and use the material after it is placed on the web.

If AI accuracy improves but customers find the product less appealing, both outcomes did not improve. If customer response improves but AI answers remain unchanged, recognize the customer outcome separately. Identifying which response changed and where shows what to revise next.

Choose one campaign and first check whether use cases and product pages support the expectation it creates.

What remains untested

We have not established which combination of material best supports AI recommendations, whether any benefit justifies production cost, or whether the same feeling carries across cultures. This was not selected as the most frequent topic in a month of observations. It is a method to test, not a proven instruction.