How to get your brand mentioned in ChatGPT

Step 1: Find the pages ChatGPT already cites

ChatGPT builds its shortlist from what it already knows plus, with web search on, a handful of pages it reads for that question. Those pages are the part you can change.

Write down the questions a buyer asks before they buy from you, in their words: your brand name, your price, you versus a competitor, and the open questions such as "best tool for" your category. Ask ChatGPT each one with search on and copy every source it links. Those pages are the raw material of the answer. If you are not on them, the answer has nothing to name you from.

When we asked ChatGPT, Claude and Gemini six questions about our own category on 21 October 2026, one run each, ChatGPT cited 0 to 4 sources per answer, Claude 0 or 9 and Gemini 8 to 13; one ChatGPT answer and one Claude answer cited nothing. For the four category questions (best tool, tools for agencies, two alternatives questions), almost every source was one of four page types: "best X tools" lists, "X alternatives" pages, "X vs Y vs Z" comparisons and glossary pages that define the term. Most of those were written by vendors in the category, plus a few independent publishers, Reddit threads, a YouTube video and a Wikipedia entry.

One run per engine is a sample, not a measurement. Read the page types as a pattern of what gets cited, not as a rate.

mentionedin.ai stores the cited sources of every answer it collects and ranks the cited domains per engine, split into your own site and third party sites. Its source gaps list shows the third party pages that engines cite, that name your competitors and do not name you, with the prompts and answers that cited each one.

Step 2: Get onto the list and comparison articles

Being named on pages the engine already trusts is the most direct route into its answer.

Claude called third party mentions "probably the biggest lever" when we asked it how a brand gets mentioned in ChatGPT, and Gemini gave the same advice: find the exact articles that already appear for your prompts and work on those. Start at the top of your list from step 1, ranked by how often the page is cited.

• A cited list or comparison article: write to the author with one paragraph on what you do, who it is for and your price, and offer a fact they can quote.
• A review site the answers lean on, such as G2, Capterra or Trustpilot: claim or create your listing and fill in the category and pricing fields.
• A cited Reddit or forum thread: answer the question that was asked, as yourself, and say you work on the product.

Be wary of anyone who sells a guaranteed mention inside ChatGPT's answer. The answer is generated per question and changes between runs, so nobody outside the engine can promise it. mentionedin.ai sends nothing to these sites: it shows you which pages to work on, and your team decides what to do about each one.

Step 3: Publish your own comparison and list pages

Give the engine a page on your own domain for each question buyers ask about your category.

The page types that get cited from third parties work on your own site too. Build one page per buyer question: you versus each named competitor, alternatives to the best known tool in your category, best for a specific niche, and a plain definition of the term your category uses. A narrow niche is easier to own than a broad one, because fewer pages compete for the answer.

• Answer the question in the first paragraph, then explain. No warm up intro.
• Put a comparison table near the top. Prices, plans, engines, limits: facts an engine can lift into its answer.
• Write down your price, even a range. A page with "contact us for pricing" leaves the engine to guess or to name a competitor that lists one.
• Show your method and say where a competitor is the better pick. In our sample Claude opened one answer by noting that almost every "best of" list it found was written by a vendor that ranks itself first. A list that admits its bias reads differently.
• Add something only you know: your own data, a test you ran, a real number from your product. Generic advice is already in the model.

Step 4: Make your entity unambiguous

ChatGPT has to know which company you are before it can recommend you.

Use the same name, the same one line description, the same category and the same pricing on your website, your LinkedIn page, Crunchbase, Wikidata if you have an entry, and every review site. Gemini put it plainly in our sample: conflicting information across the web confuses the model and leads to it skipping your brand. On your site, link those profiles from your Organization schema with sameAs, so the engine can tie them to one entity.

Name collisions are real. In the same sample, one engine answered a question about an AI visibility tool with a list of meeting transcription apps, because the two product names are close. If your name resembles another company's, say what you are in the same sentence as your name, every time.

mentionedin.ai counts a mention only when the answer means your company: your domain, a fact from your truth file or a check of the answer confirms it. An answer that uses your name for someone else is not counted and is shown to you as a mix up. Answers to your branded and pricing prompts are also checked against the facts you set, shown as how many of the checked answers matched them.

Step 5: Fix the first 50 words of every key page

Who you are, what you do, where you work and what it costs, before anything else.

An engine that reads your page to answer a question looks for the facts first. A homepage that opens with a question or a promise makes it dig. A homepage that opens with the four facts hands them over.

Do the same on your pricing, comparison and product pages, and keep the price you write there identical to the one on review sites. The mentionedin.ai page grader checks whether a page's first 50 words carry who, what, where and price. For a page that fails, one click drafts a new opening from your truth file, where every number must come from your truth file, current opening or brand profile, shown before and after so you can copy it into your site.

Step 6: Let AI crawlers read your pages

A page an engine cannot fetch, or that arrives empty without JavaScript, cannot be quoted.

Pages behind a login, a cookie wall or a script that loads the text later are the usual reason a good page never shows up. Fetch your page without JavaScript and read what comes back. If your own words are not there, fix that before anything else in this step.

The mentionedin.ai page grader runs these checks on any URL of your own domain: whether robots.txt lets GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Google-Extended and PerplexityBot read the page (for Claude it checks ClaudeBot; check Claude-SearchBot and Claude-User yourself), whether JSON-LD is present and valid in the HTML a bot receives, whether llms.txt exists, how many words are readable without JavaScript compared with the rendered page, and whether the title and meta description name your category.

Step 7: Test each prompt 3 times, then test again after every fix

ChatGPT gives a different answer to the same question from one run to the next. One answer proves nothing.

ChatGPT said it about itself in our sample: responses vary, and one test does not establish how often a brand will be recommended. So keep a fixed set of prompts (Claude suggested 20 to 50 prompts, Gemini a set of 15 to 20) and ask each one several times per engine. Then you have a rate you can compare with next month, not a screenshot.

Check whether the engine searched. An answer written from the model's memory will not change when you publish a page; it moves only when a new model version ships. An answer built from live search can change once the pages it reads change. mentionedin.ai records on every run whether the engine searched the web.

Write down every change with its date, the page and what changed, then run the same prompts again and compare. In mentionedin.ai, marking a playbook item done writes a dated entry to the fix log, and the report puts the fixes shipped next to the latest results. Two control prompts that no fix targets track how much the engine itself drifts, so you can tell a general shift from a move on the prompts you worked on.

ChatGPT also sends visitors who click a cited link with utm_source=chatgpt.com in the URL, so the traffic side shows up in your analytics without any extra setup.

Run the tests with mentionedin.ai

It finds the cited pages (step 1), shows when an answer mixes you up or gets your facts wrong (step 4), grades and rewrites your openings (step 5), runs the crawler and schema checks on your pages (step 6) and runs the tests (step 7). Getting onto other sites and writing your pages stays your team's work.

mentionedin.ai tests your prompts on ChatGPT, Claude, Google AI Overviews and Gemini, 3 runs per prompt on every engine. The Founder plan is $99 a month with 15 tracked prompts plus 2 control prompts and 204 consistency engine runs a month, enough to test all 17 fully once a month at 12 runs each, plus the page grader, the source gaps list, a playbook and a fix log. It starts with a 14 day trial; a card is required and nothing is charged before the trial ends.

We track our own four brands publicly as case studies, with today's numbers on each page, so you can see what the method says about brands we own before you trust it with yours. It does not track Perplexity, it does not write full articles, and it does not publish to your site: you copy what it gives you and ship it.

Questions people ask