Reviews help only where that chatbot already looks

You collected more stars, and the answer still named a rival, because that chatbot never opened the site where those reviews live.

Updated 7 October 2026

You asked Perplexity for a restaurant near the station and it named three places you already saw on every list. You asked ChatGPT about your product category and a competitor came back with Yelp or a magazine roundup in the sources.

You want more reviews, more press, more guide mentions. Fair — but corroboration in how shortlists get built is not “more stars everywhere.” It is enough independent pages that repeat the same name, and the assistant has to be able to open those pages for the question someone actually asked.

Unlike planted sources, these are changes you can make without buying a thread or rewriting your homepage first.

If you run a place

For a shop, the product page and the feed carry the facts. For a place, use the directory the chatbot already used for that ask, because the directory differs by company (AI place answers come from different directories).

What you can do without treating “AI” as a separate channel:

Keep one name, one address, and current hours on every directory your category uses where you are listed — Google, Apple, Bing, Foursquare, Yelp, Tripadvisor as applicable.
Let people who actually visited leave a review on the listing that showed up in the answer, when platform rules allow it. Google and Yelp both penalize paid or gated review campaigns; this is not “email every guest for five stars.”
Say on those listings who you are for, so a constrained ask has something to match — the same constraint fit logic as on your site.

More reviews on a directory ChatGPT never opened for that ask will not move that answer.

If you sell a product

Independent agreement here usually means roundups, comparison write-ups, and review sites — not your own product page restated in the assistant’s voice. That failure mode is the fake deodorant case: first-party marketing treated as though someone had already checked it.

On a blank category ask, retrieval often surfaces pages that already agree on the same incumbent — the same pattern as ask for a white t-shirt and you get Uniqlo, where opening the web can confirm corroboration rather than widen the pool. You cannot win a generic “best X” ask on merit alone when every roundup already names the same brands.

A smaller name still has an opening when the ask is specific and that constraint appears on the product page, the feed, and anywhere the assistant reads — not when you collect stars in isolation.

Press, guides, and consistent mentions

A city guide, a tourism board page, a trade publication, or a “best of” list counts when it is independent and when the assistant fetches it. Much of the press that would count as a check has told AI crawlers to stay out. Researchers at Saarland University found that about 60% of reputable news sites blocked at least one AI crawler, against about 9% of misinformation sites, as of May 2025 (arXiv:2510.10315). A headline on a blocking newsroom may never enter a live answer even if people saw it.

Planted community pages are a separate case in Companies use Reddit to change what AI recommends.

Official tourism listings and earned guide mentions matter for experiences and travel for the same reason. They are third-party pages that repeat your name in a retrievable place. They are slow, uneven, and regional. They are not a substitute for entity clarity or for fixing access when the bot never opened you at all.

What this does not do

Earned corroboration does not replace a blocked crawler, a wrong feed price, or a competitor’s manufactured thread. It does not guarantee a name on tomorrow’s answer. ChatGPT, Perplexity, and Gemini do not share one review index — log each tool separately if you measure at all.

No vendor study on this site proves that asking customers for more stars raises how often you get named rather than cited. Treat star campaigns as reputation work on the surfaces that already appear in your category’s answers, not as an “AI visibility” tactic with a score attached.

Check before you spend

Pick one ask where you were skipped. Open the answer again with browsing on if the product allows it. Write down the domains it cited — review site, directory, guide, your site alone, Reddit.

If the only source is your homepage, you do not have corroboration yet. If the sources are Yelp and a local guide, your next step is those listings and that guide footprint, after access and identity are already clean — the order in find out why you were skipped.