Guide
How to get your brand mentioned by ChatGPT and Perplexity
Seven steps, in order. Skipping straight to step three is why most attempts fail. Every step is free to do yourself; further down, what a tool costs a month, how the options compare, and the honest limit on what we measure — Perplexity, and not ChatGPT.
Updated 20 October 2026 · 8 minute read
How AI answers are built
There is no submission form for AI answers. Assistants build them by retrieving from the open web and deciding which sources to trust. So the work is not persuasion, it is making your brand the most quotable, verifiable option for a specific question.
1. List the questions buyers actually ask
Not keywords — questions, in your buyer’s words. Ten to twenty of them, spread across four intents:
- Discovery: “best accounting software for a small café”
- Comparison: “X vs Y for coffee shop bookkeeping”
- Price: “how much does café accounting software cost per month”
- Reputation: “is X any good”
If your list is full of your own product vocabulary, it is wrong. Buyers do not search for your feature names.
2. Ask the engines and read the answers properly
Put every question to ChatGPT, Perplexity and Claude, and record three things per answer: whether you are named, where you appear among the brands named, and which sources are cited. Run each question more than once — answers vary between runs, and a single miss is not a trend.
Read the citations especially closely. They tell you which domains the engines trust in your category, which is the shortcut to step six.
Record four things per answer and nothing else. Anything more and you will stop doing it by week two:
| Record | Why | What it changes |
|---|---|---|
| Were you named | The only binary in the whole exercise | Which question becomes a page in step three |
| Who was named instead | Tells you who the engine currently considers the safe answer | What your page has to be better at than theirs |
| Where you came among the brands named | Being listed fourth is not the same as being listed first | Whether you need a new page or a better one |
| Which sources were cited | The most useful column, and the one people skip | Your entire step-six list |
Two cautions about reading a single run. Engines are non-deterministic, so one miss is not a trend — we measured how much moves between two runs of the same questions seventy seconds apart in how often to check your AI visibility.
And never track a question containing your own brand name: the engine is told the name and hands it back, so the mention is guaranteed rather than earned.
3. Publish a direct answer page for each question you lose
One page per lost question. Title it with the question. Answer it in the first two sentences, then support the answer. No preamble about your mission.
The test: if someone read only the first 40 words of your page, would they have the answer? If not, an assistant will quote whoever passes that test instead.
Here is the same page opening, failing and then passing that test. The question is “how much does AI visibility tracking cost per month”, and rather than invent a fictional company we have used our own, so you can check both columns against our page on what AEO costs.
| Not quotable | Quotable |
|---|---|
| “At Bungad we believe every brand deserves to be found. Our flexible plans are designed to grow with you, with transparent pricing tailored to your needs — get in touch and we will build you a package.” | “Bungad AEO costs $99, $299 or $799 a month, month to month, with no setup fee. The $99 Starter plan covers one brand, 25 tracked questions and 4 manual checks a month on top of the scheduled one, on Perplexity, the only engine we hold a key for. Below is what changes at each price.” |
The second one contains a sentence an assistant can lift and attribute. The first contains nothing an assistant could be wrong or right about, so there is no reason to cite it.
Note what made the difference: a number, a range, and a scope — not better writing. How to write a page an answer engine will quote is the long version.
4. Put the specifics on your own domain, in text
Prices, integrations, specifications, coverage areas, support hours. Every fact an assistant would need to recommend you confidently, written as text on a page it can read. Not in an image, not in a PDF, not behind a tab that only loads on click.
This is where a brand can quietly lose. If your price is not stated, the engine quotes a review site’s estimate, or a competitor’s comparison table — and cites them.
5. Add structured data that matches the page
Organization, Product, Offer, FAQPage, Review. Schema does not manufacture authority; it removes ambiguity about what your page asserts. Keep it truthful and matched to visible content — mismatched markup is worse than none.
6. Earn mentions on the sources engines already cite
From step two you know which domains appear in your category’s answers. Get an accurate, complete presence on them: review platforms, directories, industry roundups, local press. One listing on a source the engines already quote beats five more posts on your own blog.
Be honest about what you can influence: you can correct and complete a listing, you cannot buy a recommendation. Anyone selling guaranteed placement in AI answers is selling something they do not control.
7. Re-measure monthly, and keep fixing the worst question
Models update, competitors publish, answers move. Re-run the same questions monthly and watch mention rate, citation share and position. Then fix the single worst-performing question rather than spreading effort evenly.
What this looks like in practice
When we ran our own first check, we were named in zero of eight answers. The most useful output was not the score — it was the citation lists underneath. Across those eight questions Perplexity cited 90 sources from 74 different domains in run A, of which 65 appeared exactly once (run B: 75 domains and 66).
There is no small club to join: the engine was reaching for whichever page most directly answered the question in front of it, including plenty of vendors’ own pages. You get findings like that from reading answers, never from a keyword tool.
What does not move the number
Worth saying explicitly, because each of these is sold as AEO somewhere and each of them is a way of spending a month without changing an answer.
- Volume for its own sake. Twenty posts around a topic lose to one page that answers the question in its first two sentences. Retrieval picks a passage, not a library.
- Keyword density. There is no ranking function here to game. The engine is choosing what to quote, and repetition does not make a sentence more quotable.
- Structured data that outruns the page. Markup asserting Q&A a reader never sees is worse than no markup: it is ineligible for rich results and it is a claim the page cannot support. Match it exactly or drop it.
- llms.txt on its own. Worth adding, costs nothing, but no engine has documented reading it — what llms.txt is, and whether it works is honest about the evidence.
- Blocking the crawlers and hoping. If you disallow the AI search crawlers, you are asking not to be read for search: OpenAI says an opted-out site will not be shown in ChatGPT search answers, though it can still appear as a navigational link, and Anthropic says a block may reduce your site’s visibility. That can be the right call, but it is a decision, not a default — see AI crawlers and robots.txt.
How long it takes
Publishing and structured data land within days to a couple of weeks, as assistants re-crawl. Third-party sources take longer, and can take a month or two. Anyone promising results in 48 hours is describing something other than how retrieval works.
| Step | Realistic lag | Why |
|---|---|---|
| Answer page published | Days to a couple of weeks | It has to be crawled, then chosen as a source |
| Structured data added | Same crawl | It removes ambiguity; it does not create authority |
| Specifics put into text | Same crawl | The engine stops needing a third party to state your price |
| Third-party listing corrected | A month or two | Their publishing cycle, then the engine's |
| A measurable move in the score | Two to three monthly runs | One answer flipping on a small question set is inside the noise |
That last row is the one to hold on to, and what counts as a good score has the arithmetic behind it. Judging this work after one re-run is how people conclude it does not work — does AEO actually work sets out what is proven and what is not.
What it costs a month, and what our price does not cover
All seven steps above are free to do by hand, and for a short question list that is genuinely the right answer — the free version sets out the morning’s work. A subscription is worth paying for when the list is too long to do by hand or when nobody on your side will write the pages.
Our plans are $99, $299 and $799 a month, month to month, no setup fee, with a free check first. Free and Starter use one engine. Growth and Scale use every engine we hold a key for.
Today we hold one key, so every plan uses one engine, Perplexity. Today the tiers differ by brands, tracked questions, manual checks and work done by a person — what AEO costs per month has the full breakdown.
None of those prices buys ChatGPT measurement. We do not hold an OpenAI key, so there is no ChatGPT tier, no ChatGPT add-on and no figure we could sell you for it at any price. The adapter is written and waiting on a key; the Claude and Gemini adapters are in the same state.
Google AI Overviews and Microsoft Copilot have no public consumer API, so any vendor’s number for those two came from scraping a results page or from a person reading a screen. Everything on this page about how to earn a mention applies to ChatGPT — it is a retrieval engine like the others — but the measurement we sell does not cover it.
If ChatGPT numbers are what you need this month, the honest advice is to buy them elsewhere: a self-serve tracker that holds the key will have them, and the tools compared names which, with each entry price dated and sourced.
The free and first-party alternative for Copilot is the AI Performance report inside Bing Webmaster Tools, which you can open yourself without any vendor.
On whether we are any good at this: our own audited visibility score is 0 out of 100, measured on the eight questions our own buyers ask, published with the whole run and the competitors named instead of us. We know of no other vendor in this category that publishes theirs.
Judge a provider on that, on which engines it automates rather than eyeballs, and on whether it will put both answers in writing — the seven questions to ask an AEO vendor is that interview, including the four we lose.
Common questions
How much does it cost to get mentioned by ChatGPT?
Nothing, if you do the seven steps above yourself — there is no paid inclusion in an AI answer and no placement to buy. If you want the work managed, our plans are $99, $299 and $799 a month, month to month, with a free check first. Be clear about what that buys: every plan is one engine, Perplexity, because that is the only key we hold. No plan of ours measures ChatGPT at any price, the adapter is written and waiting on a key, and if ChatGPT numbers are what you need this month you should buy a tracker that holds the key instead of us.
Can I pay to be included in AI answers?
No. There is no submission form, no paid inclusion and no placement to buy. Assistants build answers by retrieving from the open web and choosing which sources to trust, so the only levers are being findable, being quotable and being corroborated elsewhere. Anyone selling guaranteed placement is selling something they do not control.
How long before my brand appears in AI answers?
Publishing and structured data land within days to a couple of weeks, as assistants re-crawl. Third-party listings can take a month or two. A move large enough to read on a score can take two to three monthly runs, because a single answer changing its mind on a small question set is inside the noise.
Do I need to publish a separate page for every question?
For every question you lose, yes, and only for those. One page per question, titled with the question and answering it in the first two sentences. Bundling five questions into one long page means an engine has to decide which passage answers the question, and it will usually find a page elsewhere that made that decision for it.
Which matters more, my own pages or third-party sources?
Your own pages come first, because they are the only thing you control and because specifics an engine cannot find on your site get quoted from somewhere else. Third-party sources matter too, though: in our own audit a competitor’s own domain was among the most cited sources, and so were publishers and directories rather than vendors.
Should I track a question that contains my brand name?
Ask it, but never count it. An engine asked what reviewers say about your brand is told the name and hands it back, so the mention is guaranteed by the question rather than earned and it can never be lost. It is worth reading and worthless as a measurement.
More guides
Read next
Want a short version of the first two steps done for you?
Enter your domain. We ask Perplexity three questions your customers ask and show you whether you were named and which domains the engine cited.
3 questions on Perplexity, free. A domain checked in the last week is served from that stored run rather than asked again, and the daily free allowance resets at midnight UTC.
