How to Get Mentioned in ChatGPT in the Next Two Weeks

Why do AI assistants mention some businesses and ignore others?

AI assistants mention the businesses they can identify, verify and quote in a single sentence. That's the whole game. The model has to know who you are as an entity, find independent confirmation elsewhere on the web that you exist and do what you say you do, and then lift a specific line from a page that answers the question being asked. Miss one of those three and you stay invisible, however handsome the website is. Most owners assume ChatGPT reads their homepage and forms an opinion. It doesn't. It assembles an answer from whatever it can retrieve and trust in about two seconds.

Two different mechanisms are at work, and they reward different things. When ChatGPT answers without browsing, it's drawing on training weights — it either learned your brand already or it didn't, and no amount of last-minute publishing changes that this week. When it searches, the prompt gets broken into several shorter queries, a pool of candidate pages comes back, and a smaller model reranks them for relevance and reliability before picking a handful to cite. The first path rewards being a known entity. The second rewards having a crawlable page with a clear, checkable claim on it.

Here's the part that surprises people: when an AI recommends your business, the link it shows often isn't yours. It's a review platform, a directory, a roundup post or an industry publication that happens to name you. Models treat agreement across independent sources as evidence, so being described consistently in three places you don't own beats describing yourself beautifully in one place you do. That's uncomfortable if you've spent years polishing your own site, and it's the single biggest reason two similar businesses get wildly different treatment from the same assistant.

What do ChatGPT, Gemini and Perplexity actually need from your website?

They need permission to fetch your pages, text that exists in the HTML, and an unambiguous statement of who you are. Permission comes first, because it's the one lever that's officially documented and the one most sites get wrong by accident. OpenAI runs three separate crawlers with three separate jobs, and they can be allowed or blocked independently in robots.txt. Plenty of sites block all of them in one line without realising they've just opted out of being cited. Worse, some CDNs and security layers now block AI crawlers by default, so your robots.txt can say yes while your firewall quietly says no.

Once the door is open, make the identity obvious. Organization schema with your legal name, address, phone number and logo, plus sameAs links pointing to the profiles that represent you elsewhere, gives the engines a clean way to connect your website to a known entity rather than guessing. Keep the details identical everywhere — same business name, same suite number, same phone format. Inconsistency doesn't just weaken the signal, it splits you into two half-confident entities. Add publish and update dates to articles, and render your important text server-side so it's there without JavaScript.

The engines then behave differently, and it's worth knowing how. Google's AI Overviews and AI Mode are grounded in Google's own index, so they reward a passage that supports a claim; the Gemini app leans harder on whether it recognises your brand at all. Perplexity is aggressively fresh and citation-hungry — it'll happily cite a page published last week and stack many sources into one answer. Claude cites sparingly and favours curated review and listicle pages. One site, one set of fundamentals, but the payoff arrives on different surfaces at different speeds.

Which pages and formats earn AI citations most reliably?

Four page types earn AI citations far more reliably than anything else: direct-answer pages that resolve one specific question, comparison pages that weigh real options against each other, pricing and process pages with concrete numbers, and genuine FAQs written in the words customers actually use. What they share is extractability. Each one contains short, self-contained passages that can be pulled out and quoted without the surrounding context. A rambling thought-leadership essay may be more enjoyable to read, but it gives a reranking model nothing clean to grab, so it loses to a plainer page with a sharper answer.

Structure carries more weight here than style. Put the question in the heading, then answer it in the first two sentences underneath — declaratively, with the specifics attached. "Most clinics in this range charge between X and Y, depending on Z" beats "pricing varies according to a number of factors." Name the numbers, the timeframes, the exceptions and the cases where your service is the wrong fit. Honesty about limitations is oddly effective: it produces the kind of concrete, checkable sentence that retrieval systems like, and it reads as trustworthy to the human who lands on the page afterwards.

Now a contrarian note on llms.txt. It's cheap to publish and it forces you to think about your own information architecture, so go ahead and ship one. But don't let anyone sell it to you as the key to AI visibility. No major provider has committed to reading it in production, Google has said plainly that it doesn't support the format, and server logs from people who've measured it show AI crawlers fetching ordinary HTML almost exclusively. An hour on llms.txt is fine. A quarter spent on it instead of content and crawl access is a quarter wasted.

And the strategic point underneath all of this: every one of these pages keeps working after you publish it. That's the difference between organic marketing and paid ads. A campaign stops delivering the day the budget stops. A well-built comparison page can be cited by an assistant eighteen months from now, to someone you'll never meet, at no extra cost — and ads can't appear inside an AI answer at all.

What should you do in the next two weeks to get mentioned in ChatGPT?

Spend days one to three on access and identity, because nothing else works until those are right. Open robots.txt and check each AI crawler by name rather than as a group. Check your CDN or security provider separately, since its default settings may override your intent. Confirm your key pages are indexed in Bing as well as Google — ChatGPT's search layer draws on a blended index, and being missing from Bing is a common, silent handicap. Then add or correct Organization schema and make sure your business name, address and phone match everywhere they appear.

Days four to nine belong to content, and three or four sharp pages beat fifteen vague ones. Write one direct-answer page for the question your customers ask most; one comparison page that honestly weighs your approach against the obvious alternative, including who should pick the alternative; one pricing or process page with real ranges and real timelines; and one FAQ built from the questions your inbox actually receives. Keep each answer self-contained under its heading. Date everything. Resist the urge to pad — padding is exactly what gets a passage passed over.

Days ten to fourteen go off-site, and this is the work most businesses skip. Claim and complete the directory and review profiles that matter in your category and your city, with descriptions that match your site word for word on the key facts. Get yourself into two or three roundups, association listings or partner pages where a human editor names you. Answer questions in the forums and communities where your buyers already argue about this stuff — not with a pitch, with a useful answer. Independent mentions are the raw material AI engines use to decide you're real.

Two weeks won't make you the default recommendation in a competitive category. What it will do is remove the blockers, give the engines something worth quoting, and start the clock on the slower part — accumulating mentions. Expect early movement on Perplexity, which refreshes fastest, then Google's AI surfaces, then ChatGPT's underlying sense of who you are, which is the slowest to shift. Treat the fortnight as the foundation, then keep publishing at a steady rhythm rather than in bursts.

How do you know it's working, and what does AI share of voice tell you?

AI share of voice is the percentage of AI answers that mention your brand out of all brand mentions across a fixed set of prompts — your mentions divided by everyone's mentions, times one hundred. It's the closest thing this field has to a rank tracker, and unlike rankings you have to generate the data yourself, because no assistant publishes a search console. The method is unglamorous: build a prompt list, run it on a schedule, record who gets named, and watch the trend rather than any single answer.

Build a panel of roughly twenty-five to fifty prompts, and weight it heavily towards discovery questions — the ones where a buyer describes a problem and names nobody. Asking ChatGPT what it thinks of your company proves almost nothing; your name is already in the conversation. Run the panel across ChatGPT, Gemini, Perplexity and Claude, run each prompt at least twice because answers vary between attempts, and count each brand once per answer no matter how often it repeats. Record four things: mentioned, cited with a link, recommended as a top pick, and the sentiment attached.

Keep the panel frozen. Change the prompts and the schedule at the same time and your trend line becomes unreadable. A full run monthly, plus a weekly spot-check on your ten most commercially important prompts, is a sensible cadence for most businesses, and it also catches assistants repeating outdated prices or discontinued services about you. If you'd rather see where you stand before committing to the whole routine, a quick GEO score check is a reasonable starting point. And if you want the research, writing, publishing and tracking handled on autopilot, book a demo and we'll walk you through it.

Frequently asked questions