Do First-Party Review Pages Change AI Answers? Test Protocol
A consent-based experiment for testing whether a crawlable first-party review page is retrieved or reflected in repeated AI answers. Results pending.
Protocol published August 13, 2026; testing pending. This experiment will ask whether a new, permissioned first-party review page is retrieved or reflected in repeated AI answers about one consenting local business. It does not currently claim that testimonial pages change ChatGPT, Gemini, or any other answer system.
A local-SEO discussion reported a before-and-after change after adding testimonials. That anecdote supplies a reader question, not controlled evidence. Model updates, index changes, third-party sources, account state, location, and ordinary response variance can all change an answer.
Primary question
After a crawlable first-party review page is published with no other material site change, does the page appear in retrieved sources or do its permissioned claims appear more often in a predeclared set of repeated answers?
Ethics and eligibility gate
- The business owner consents to the study and can stop it.
- Every testimonial is genuine, permissioned for publication, and accurately attributed or anonymized.
- No review is purchased, fabricated, selectively rewritten, or marked up as an independent third-party rating.
- Personal data is minimized; removal and correction requests have an owner.
- The page visibly explains how testimonials were collected and selected.
Freeze the baseline
| Variable | Record | Reason |
|---|---|---|
| Prompt | Exact text and order | Prevents question drift |
| Surface | Product, model when shown, account state | Answers differ by surface |
| Context | Region, language, location permission, date | Local answers are context sensitive |
| Response | Full answer, citations, links, screenshots | Preserves the observed output |
| Claim coding | Business facts, sentiment, distinctive review themes | Separates retrieval from generic praise |
Run the declared prompts repeatedly before publication and save answers that omit the business or reviews. The page and site should remain unchanged during the baseline.
Treatment
Publish one accessible HTML page containing the permissioned review text, collection method, review date, service context, author-display choice, business response where appropriate, and correction contact. Link it normally from the site. Do not request indexing repeatedly or change unrelated pages to force a result.
Confirm ordinary eligibility: final status, canonical, robots controls, internal link, visible text, mobile rendering, and sitemap inclusion where appropriate. Google’s AI features guidance does not provide a special schema or guaranteed inclusion path.
Declared recheck
After a stated crawl window, repeat the same prompts under the recorded context. Code whether the new page was cited, linked, or apparently reflected; whether third-party sources changed; and how much answers vary across repetitions. Preserve null and contradictory results.
Interpretation rules
- A citation to the page shows retrieval in that response, not a permanent ranking.
- A matching phrase without a citation does not prove the page was the source.
- A before-and-after difference is an association unless competing changes are excluded.
- Generic positive wording is not coded as review influence.
- The result applies only to the business, prompts, surfaces, region, and dates tested.
OpenAI’s publisher guidance explains crawler and referral controls, but access does not guarantee inclusion or citation. Use SEA’s citation-ready passage test to make claims portable, without treating it as a control over an answer engine.
Community question
The protocol was prompted by a discussion in r/localseo. The anecdote is not used as evidence that the treatment works.
Ask a question or join the discussion