AI search
How do I get my business mentioned by ChatGPT, Claude and Gemini?
An assistant can only name your business if it can read your pages without running JavaScript, if those pages answer questions directly rather than teasing an answer, and if the places it sources local results from — Google Business Profile, Yelp, Foursquare, Facebook and Reddit — agree about who you are. Nobody can guarantee you a position: ask Gemini the same question twice and it names the same top business only about 7% of the time. What you can buy is eligibility.
The short version
- Step one is technical, not editorial: if the page is empty without JavaScript, nothing else matters.
- Assistants quote pages that answer the question in the first forty words.
- ChatGPT now sources 70%+ of local results from Foursquare, and OpenAI licensed Yelp’s 330M reviews in July 2026.
- AI referrals are about 1.08% of site traffic today. Treat it as a compounding side bet funded by ads, not an ads replacement.
How to do it
Prove the page is readable without JavaScript
Fetch your own homepage the way a crawler does and look at what comes back. If it is an empty shell, every assistant sees an empty shell. This is the single most common disqualifier and it is invisible in a browser.
Restructure pages answer-first
The H1 is the question. The first paragraph is the complete answer, written so it stands alone if someone copies it out. Everything after that is the reasoning. Pages built to force a click get skipped.
Publish the four pages buyers actually ask about
What it costs, what goes wrong, who it is not for, and repair versus replace. These are the questions typed into assistants, and almost no local business has a page for any of them.
Fix the sources the models read for local results
Google Business Profile, Yelp, Foursquare, Apple Business Connect and Facebook. Same name, same address, same phone, same categories. Disagreement between them is read as uncertainty about whether you exist.
Get reviews that contain sentences, not just stars
Gemini matches on review text rather than star count, so a review saying "showed up on time and cleaned up after" is worth more than a silent five stars. Ask for the specific thing you want quoted back.
Add llms.txt and keep it honest
A plain-text file at your domain root stating who you are, what you sell, what it costs and where the canonical pages live. It is not magic and no crawler is obliged to read it — but it costs nothing and it removes ambiguity for the ones that do.
The disqualifier nobody checks
We crawled the homepages of 1,348 companies that had just raised funding and measured how much of each page survives without JavaScript. 8.7% returned a shell. The median broken one served 55 characters of readable text where the rendered page had 5,358.
The counterintuitive part is who was affected. Sites built on modern JavaScript frameworks were the worst — Next.js at 8.5%, HubSpot CMS at 10.0%. Sites on Wix and Squarespace produced zero failures out of 91. A company that raised millions and hired real engineers was more likely to be invisible to ChatGPT than one paying $16 a month for Wix.
This is not an argument for Wix. It is an argument for checking, because the failure is completely invisible to the person who built the site — it renders perfectly in their browser.
What "answer-first" actually means
Most business pages are written to create curiosity. An assistant does not experience curiosity. It extracts whatever passage best answers the question and attributes it.
So the page needs a passage worth extracting: a direct answer, early, that makes sense with no surrounding context. If your pricing page says "every project is unique, contact us for a quote", there is nothing to extract and you will not be named. If it says "most kitchen remodels in Metro Detroit run $8,000 to $14,000, driven mainly by cabinet choice and whether plumbing moves", that is a quotable sentence and you become the source of it.
The same structure that makes a page quotable also makes it convert better with humans, which is the rare case where the optimisation and the honest thing are identical.
Where each assistant actually gets local answers
The advice that circulates here is mostly out of date. Two things changed recently and both matter more than any on-page tactic.
ChatGPT no longer leans on Bing for business results the way it did. Foursquare now supplies more than 70% of its local results, and in July 2026 OpenAI licensed Yelp’s 330 million reviews and 8 million listings. If your Yelp and Foursquare records are wrong or missing, that is the gap.
For Bing Places, Whitespark found Facebook is the number one review source surfacing in results — ahead of Yelp. Almost nobody works Facebook reviews, which makes it one of the few genuine edges still lying around.
And Reddit remains the second-largest local citation source at 13.7%, beating Angi, Thumbtack and HomeAdvisor combined. Reduced from its peak, not gone.
Why nobody can sell you an AI ranking
Ask Gemini the same local question twice and it names the same top business roughly 7% of the time. The cited sources overlap about 40%. The output is not a ranking — it is a sample from a distribution that changes between identical queries.
Anyone promising you position one in ChatGPT is selling something that does not exist. What can be delivered is eligibility: being readable, being quotable, being consistent across the sources the models actually read. Do that and you show up some of the time instead of none of the time.
Size the opportunity honestly too. AI referrals are around 1.08% of site traffic. They convert better than average, though the published multiples are all over the place and the methodology behind them is weak. It is a compounding side bet, funded by the channel that pays the bills today.
| Layer | Effort | What it changes |
|---|---|---|
| Server-rendered / prerendered pages | One-time engineering | Eligibility. Without it nothing else can work. |
| Answer-first page structure | Ongoing writing | Whether you get quoted once you are readable. |
| Listing consistency (Google, Yelp, Foursquare, Apple, Facebook) | A day, then quarterly | Whether local answers name you at all. |
| Review text, not just star count | Ongoing ask | Which attributes get attached to your name. |
| llms.txt and structured data | A few hours | Removes ambiguity. Marginal, but free. |
Common questions
- Does llms.txt actually do anything?
- No crawler is required to read it and none has documented doing so as a ranking input. It takes an hour, it costs nothing, and it gives any agent that does fetch it an unambiguous statement of what you sell and what it costs. Treat it as cheap insurance, not as a strategy.
- Should I block AI crawlers to protect my content?
- If you want to be cited, no — blocking GPTBot, ClaudeBot, PerplexityBot or Google-Extended removes you from the answers. Blocking makes sense if your content is the product and being summarised destroys its value. For a local service business it is self-harm.
- How do I know if my site is readable by AI assistants?
- Request the page the way a non-JavaScript crawler does and read what comes back. If the visible text is under a couple of hundred characters, your page is a shell. We run a free check at revioconsulting.com/tools/ai-visibility that does exactly this and shows you the text a crawler receives.
- Is this replacing SEO?
- It is the same work with a different acceptance test. Readable pages, clear structure, real answers, consistent listings and genuine reviews were always the fundamentals. What changed is that the reward moved from a blue link position to being the quoted source, and that pages written to withhold the answer now lose twice.
Where these numbers come from
- 8.7% of 1,348 funded-company homepages were JavaScript shells; Next.js 8.5%, HubSpot CMS 10.0%, Wix and Squarespace 0%.
- — Revio, The Funded Site Audit, September 2026
- Foursquare supplies 70%+ of ChatGPT local results; OpenAI licensed Yelp’s 330M reviews and 8M listings on 2026-07-23.
- — Reported platform sourcing changes, 2026
- Facebook is the number one review source surfacing on Bing Places, ahead of Yelp.
- — Whitespark, 153 queries across 17 categories and 9 cities
- Reddit is the second-largest local citation source at 13.7%.
- — Whitespark local citation analysis
- An identical Gemini query names the same top business only about 7% of the time; cited sources overlap about 40%.
- — Repeat-query grounding drift testing
- AI assistants refer approximately 1.08% of website traffic.
- — Conductor
Last reviewed .
Read next
Want this done for your business?
Tell us where you are trying to get to. If the honest answer is that you do not need us, we will say so.