How to get recommended by ChatGPT: the 2026 method

A worked example: 1 mention out of 12, then 5

Northline sells scheduling software to dental clinics. On 2 November 2026 the founder asked ChatGPT for the best scheduling software for a two-chair clinic. Four products came back. Northline was not one of them. Getting into that list is the work, and how to get recommended by ChatGPT is the question Northline spent the next five weeks on.

The founder wrote eleven more questions a clinic manager would type with no brand in mind, then checked all twelve by hand, signed out, on ChatGPT, Gemini, an AI Overview and Google AI Mode. Northline appeared in one answer out of twelve, at the bottom of a list of other options. Two answers cited a directory page showing a price of $79 per month. The real price was $49.

Five weeks later the panel score was 5 out of 12. Northline was named third for the two-chair question, with a link to its own pricing page. The software did not change. Three things did: the pages the answers were built from, the prices on third-party listings, and the sentences that tied the product name to the need it answers.

What happens between the question and the answer

A buying question turns into one or more web searches. ChatGPT searches Bing and OpenAI's own crawl. AI Overviews and AI Mode search Google. A page that does not surface in those searches cannot end up in the answer.

The engine picks the results that look most on-topic, specific and current, writes a short answer, and lifts passages from the pages it picked. Most lifted text comes from near the top of a page.

That gives three gates: be found, be picked, be quoted. Writing quality only affects the third.

Being cited is not the same as being named

Across 115 prompts in 14 countries, 61.7% of citations in AI answers were ghost citations. The answer used the page as a source and never mentioned the brand behind it. Your page can be the reason ChatGPT recommends a competitor.

Citation is a page problem. Naming is a sentence problem. A rewrite that fixes the page will not fix the sentence.

Step 1: Write a question panel and freeze it

Twenty to fifty questions your buyers type before they have heard of you. Three kinds cover most of the ground.

Kind Example What it tests
Discovery best scheduling software for a dental clinic whether you exist in the category at all
Problem how do I stop patients double-booking whether your page answers a need people search for
Comparison cheaper alternatives to the tool every clinic uses whether you show up beside the names you compete with

Leave your own brand out of the question. Asking "is Northline good" tells you nothing about whether you get recommended.

Specific questions are where smaller brands appear. In a study of 210 buying questions across 21 categories, four engines put the same brand first on only 27% of questions, and 16% of all recommendations came from a single engine. A general running-shoe question returns the same five names every month. A question about wide feet returns different brands.

Freeze the wording and give each question an ID. Run the panel on the same day each month, signed out of every account. Record five things per answer: whether the brand is named, at what position, whether a link points to a domain you control, who else is named, and which sources the answer used. Signed-out sessions still have web search, so a clean run is realistic. Add new questions with new IDs instead of editing old ones, so month four can be compared with month one.

Step 2: Build one page per question

Not one guide covering everything. In a study of 815,000 query-page pairs, focused pages picked up ChatGPT citations more often than long "ultimate guides" on the same topic.

Each page needs:

  • A title phrased as the buyer's question
  • A first paragraph that answers it and names the options, including yours
  • A price for every option. "Contact us" gives the engine nothing to quote
  • A comparison table with specs and prices, with a source for each
  • A short section on who the product is not for
  • A visible updated date that you keep accurate

Step 3: Write the sentence you want repeated

The passage lifted from your page has to carry your brand and the need it answers in the same line.

Weak: "Our platform helps clinics manage appointments."

Strong: "Northline is scheduling software for clinics with one to three chairs. It costs $49 per month per location and sends appointment reminders by SMS."

Cited text runs at an entity density of 20.6%, against 5 to 8% in ordinary English. Brands, products, prices, numbers. A page that says "great product" teaches the engine nothing.

Two patterns show up in an analysis of 18,012 citations:

  • 44.2% of citations come from the first 30% of the page
  • 78.4% of citations tied to a question come from a heading

Write the buyer's question as an H2 and answer it in the paragraph directly below.

Step 4: Check the pages you do not own

Directories, review sites and listicles get cited heavily for commercial questions, and they go stale. The two outdated prices in Northline's baseline came from a directory nobody had touched since the previous year. Claim the listings, correct the price, and add a line that names the product and the category. A review that says "Northline is the cheapest scheduling tool with SMS reminders" teaches the engine the link between the brand and the need. A review that says "great product" teaches it nothing.

The four factors that decide whether a page gets cited

A controlled experiment ran 252,000 trials across six models and 18 page factors. Four decided whether a page was cited, in every model tested:

Factor In practice
Topic match The page answers that exact question rather than the whole category
A stated price A number in the text, dated
A recent timestamp An update date you keep honest
Rank in results Where the page sits in the search results the engine used

Missing any one of the four could rule a page out. Specs, comparisons, evidence, depth and confident wording came next. Formatting by itself had no measurable effect on whether a page was picked.

Two routes exist. The first is getting third-party pages to cite you, which takes months and depends on other people's publishing schedules. The second is the conversation itself: assistants are trained and tuned on text that includes people asking about products, getting answers, and reacting to them. Brand-positive conversation is an input you can influence directly.

GEO Boost runs brand conversations in AI assistants as behavioral signals for LLMs. A campaign opens chats in the public web interfaces of ChatGPT, Gemini, Perplexity, Microsoft Copilot, Google AI Mode, Mistral Le Chat and Yandex Alice, in a real browser, without logging in. Each chat starts with a question a real customer asks, either written by you or generated from Google keyword data. The assistant answers, and the conversation then names your brand, asks about its strengths and praises it, for a set number of messages, five by default. Every conversation leaves through a residential connection in the country you choose, and the location of each connection is verified. Countries have their own daily chat volumes, spread across the day, and follow-up messages are written in the language of the first question and adapted to the answer that came back, so no two chats read alike.

Each transcript is stored and each chat is scored: did the assistant name the brand on its own in the first answer, at which position, did it agree with the user by the end, and in what tone. The dashboard tracks the share of first answers that name the brand over time, which is the number to compare against your question panel. The start of it is free: new accounts get credit with no card. After that you pay as you go per conversation, or take the Basic plan at $99 per month for 150 conversations. Prices checked November 2026. You can create an account and open a first campaign and read the first answers yourself.

The mechanics sit in LLM behavioral signals: what user conversations tell AI assistants.

How to measure whether ChatGPT recommends you

The question panel

Score each answer from 0 to 3: 0 not mentioned, 1 mentioned, 2 linked to a URL you control, 3 named first. Run it monthly, same day, same country, signed out, on four surfaces. Write down the cited URLs and any wrong facts, because a stale price in a cited source is a fixable problem. The full scoring sheet is set out in how to measure brand mentions in AI answers.

Referral traffic

ChatGPT appends utm_source=chatgpt.com to the links it shows, and those visits appear in GA4 and in most analytics tools. Build a segment for chatgpt.com and perplexity.ai and make sure AI visits land inside it instead of being split across direct and referral. Cross-site referrers are usually trimmed to the origin under the default strict-origin-when-cross-origin policy, so you learn which engine sent the visit and never the question. Some AI visits arrive as direct traffic. Treat the count as a floor.

Search Console and Bing

Google counts AI Overviews and AI Mode inside the normal Performance report under the "Web" search type, so a rise there mixes AI answers with ordinary results. Bing Webmaster Tools has a separate AI Performance report with citations and grounding queries for supported experiences. Those counts are sampled and represent citations rather than clicks.

The primary references for all of this are Google's documentation on AI features in Search, the Performance report documentation and the HTTP Referrer-Policy reference.

Expect the answers to move

AI answers change with the wording of the question and the location of the person asking. In one study, Google's two AI products shared only 53% of their recommendations for the same question, less overlap than between ChatGPT and Gemini. A single test is one sample. Run the panel again before you believe any number, upward or downward.

A 30-day sequence

Week Work
1 Write 20 questions, run the first baseline signed out on four surfaces, record scores and cited URLs
2 Rewrite the pages behind the questions where a competitor is quoted, add prices and real update dates
3 Fix stale directory and review listings, add a sentence that names the brand and the need
4 Re-run the panel, compare with week 1, and start a conversations campaign for the questions where the text changes moved nothing

FAQ

Most page changes show up in a panel within two to six weeks, because a recent timestamp and fresh content both feed the selection step. Take the first measurement before any work starts, or there is nothing to compare against. Every baseline should be dated, and 32 questions with a written record of answers is a reasonable starting size.

Do ChatGPT ads put a brand inside the answer?

No. Ads run on systems separate from the model, are labeled as sponsored and appear below the answer. An ad campaign buys a paid slot under the answer, and the model's own recommendation above it is written the same way whether or not the ad exists.

Can anyone guarantee a mention in an AI answer?

No. Assistants word answers differently each time the same question is asked, and the brands they name shift with the phrasing of the question and the location of the person asking. What can be measured is the share of answers that name you across a fixed panel, month by month, which is why the panel matters more than any single screenshot.

How many questions should I track?

Twenty is workable, fifty is better. The panel needs enough specific questions to catch the ones where a smaller brand can appear, since general category questions tend to return the same handful of names every time.

Does being cited by ChatGPT mean my brand will be named?

Often not. In a study of 115 prompts across 14 countries, 61.7% of citations were ghost citations: the page was used and the brand behind it never appeared in the written answer. Whether your name survives depends on whether the lifted sentence carries it.

What does GEO Boost cost?

New accounts start with free credit and no card required. After that it is pay as you go per conversation, with no subscription, or the Basic monthly plan at $99 for 150 conversations. Prices checked November 2026.

More articles · GEO Boost