AI search

How to get your business recommended by ChatGPT in 2026

August 5, 2026 · 9 min read · Pedro Paucar

A customer who needs an emergency plumber at 9pm no longer types four words into Google and scans ten blue links. They ask an assistant, in a full sentence, and they get one answer with two or three businesses named in it.

That shift is now measurable rather than theoretical. In 2026, fewer than a third of Google searches send a click to any website at all, and in Google's AI Mode the figure is far starker — analysis of Semrush data suggests roughly 93% of AI Mode searches end without a click to an external site. When an AI Overview appears above the results, organic click-through rates fall by around two thirds.

The practical question for a local business is no longer "how do I rank?" It is "how do I become one of the two or three businesses the assistant names?" Here is the answer, in the order I actually implement it.

First, understand what broke

The most important 2026 finding is not the traffic decline — it is the divergence. The overlap between ranking in Google's top ten and being cited inside an AI Overview fell from roughly 75% in mid-2025 to somewhere between 17% and 38% in early 2026.

Read that twice, because it explains a confusing experience a lot of owners are having right now: their rankings look fine, their positions have not moved, and the phone rings less. They did not lose the ranking. They lost the citation. Those are now two different competitions with two different rulebooks.

The old rulebook rewarded pages that were findable. The new one rewards facts that are verifiable. A language model naming a business in an answer is making a small reputational bet, and it hedges that bet the way a cautious person would: it prefers claims it can corroborate from more than one independent place.

1. Make your business facts identical everywhere

This is unglamorous and it is the single highest-leverage thing on the list. Your business name, address, phone number and hours must be byte-for-byte identical on your website, your Google Business Profile, your Facebook page, Yelp, Apple Maps, industry directories and every listing you have ever filled in.

"Suite 200" on one and "Ste. 200" on another is not a small inconsistency to a retrieval system — it is two candidate facts with no tiebreaker, and the safest move for the model is to name someone else. Pick one canonical format, write it down, and enforce it everywhere.

Add hours, service area, payment methods and languages spoken. An assistant asked for "a contractor near Baltimore who speaks Spanish" can only answer that if somebody, somewhere, published the fact in machine-readable form.

2. Answer the question in the first paragraph

Analysis of LLM citations found that about 44% of all citations come from the first 30% of a document. Models pull from the top. Everything you buried under a 400-word introduction about "the importance of choosing the right provider" is invisible.

So invert the structure of every page:

• Lead with the answer. If the page is "how much does a bathroom remodel cost in Maryland," the range appears in the first two sentences, not in section four.
• Write one page per real question. Not "Services" — "Do you offer same-day appointments?", "What areas do you serve?", "How fast can you start?"
• Use the customer's words. People ask assistants in full sentences. Your headings should read like those sentences, because that is what gets matched.
• Include a short summary box at the top of long pages. It gives the model a clean, quotable block — and it gives the human skimming on a phone the same thing.

There is a pleasant side effect here: every one of those changes also improves the page for people. This is the rare optimization that does not trade one audience against another.

3. Add the structured data that names you

Structured data is a block of JSON in your page's code that states, unambiguously, what the page is about. It is not decoration — it is the difference between a machine inferring your hours and knowing them.

The three that matter for a local business:

• LocalBusiness — name, address, phone, hours, geographic service area, price range.
• FAQPage — your real questions and answers, in a format built to be quoted.
• Article with a named author on anything you publish — including who wrote it and what they actually do.

That last point matters more than it looks. Anonymous content is cheap and 2026 is drowning in it. A named human with a verifiable background attached to a claim is a signal that survives the flood.

4. Get mentioned somewhere that isn't your website

Here is the uncomfortable part: your own website is your weakest source about you. Everything on it is a claim you made about yourself. Models weight third-party corroboration heavily, which is why the businesses that get recommended tend to be the ones that exist in multiple places.

What counts, roughly in order of value:

• Review platforms with real, recent, detailed reviews — not just a star average.
• Local press, chamber of commerce listings and community organizations.
• Reputable industry directories and trade association member pages.
• Genuine participation in public discussion forums where your expertise is on display and useful.

One well-written local news mention outperforms fifty pages of blog content that only exists on your own domain. This is the part most businesses skip because it cannot be done in an afternoon — which is exactly why it works.

5. Make the page fast — genuinely fast

One of the more surprising findings from 2026 citation research: pages with a First Contentful Paint under 0.4 seconds are about three times more likely to be cited by ChatGPT than pages loading above 1.13 seconds.

The mechanism is mundane. Retrieval systems operate under time budgets. A page that does not respond quickly enough is a page that does not make it into the answer, no matter how good the content is.

For most small business sites the fix is not exotic: serve images at the size they are actually displayed and in a modern format, cut the third-party scripts you are not using, and stop loading fonts and trackers that block the first render. On this site the icon set alone went from 19 MB to 140 KB by exporting each image at the size it is genuinely shown. Nothing looked different. Everything loaded faster.

What about llms.txt?

You will read a lot of advice about adding an llms.txt file — a plain-text summary of your site for language models. It is trivial to add and it does no harm, so add it if you like.

But be honest about what it buys you. Research published in April 2026 found the measured effect on citation rates to be minimal. What correlated with citations was domain authority, structured data, review platform presence and third-party mentions — items 1, 3 and 4 above. Treat llms.txt as a courtesy, not a strategy.

Then measure it the only way that works

There is no Search Console for ChatGPT. So do it manually, once a month, and write the results down:

1. List the ten questions a customer would actually ask before hiring you.
2. Ask each one to ChatGPT, Google AI Mode, Perplexity and Gemini — in a fresh session with no memory of you.
3. Record whether you were named, who was named instead, and which source was cited.

That last column is the whole game. Whoever is being cited has already solved the problem you are solving, and their page is a free specification of what the model wanted.

The uncomfortable conclusion

The businesses winning AI recommendations in 2026 are not the ones with the cleverest tactics. They are the ones whose facts are consistent, whose answers are direct, whose reputation exists outside their own domain, and whose site is fast.

Which is to say: the shortcut era is over. The good news is that this work compounds and it is largely invisible to competitors who are still counting keyword density. And when a visitor does arrive from an AI answer, they arrive with high intent and low patience — which is precisely why what happens in the next sixty seconds on your site matters as much as the citation that brought them. If nobody answers, the assistant that recommended you has already offered them two alternatives.

Frequently asked questions

How do I get my business recommended by ChatGPT?

AI assistants recommend businesses they can verify from more than one source. Keep your name, address, phone and hours identical everywhere; publish pages that answer the exact questions customers ask, with the answer in the first paragraph; add LocalBusiness and FAQPage structured data; earn mentions on third-party sites like directories, local press and review platforms; and keep the pages genuinely fast.

Is SEO dead now that AI answers most searches?

No, but ranking and being cited have come apart. The overlap between Google's top ten results and the sources cited in AI Overviews fell from roughly 75% in mid-2025 to between 17% and 38% in early 2026. A business can hold its rankings and still lose traffic because it is not the source the AI quotes.

Does an llms.txt file help my site get cited by AI?

It costs nothing and does no harm, but the measured impact on citation rates is minimal. Research published in April 2026 found that what actually drives citations is domain authority, structured data, review platform presence and third-party mentions. Add the file if you want, then spend the real effort on those four.