Guides

How to get ChatGPT to recommend your business (what 839 AI answers and 6 studies show)

What 839 AI answers and six published studies show. Every claim is numbered to its source at the bottom.

When someone asks ChatGPT "what's the best CRM for a small team", they get four or five names and usually stop there. This guide covers how to become one of those names: on ChatGPT, and on Gemini and Google's AI Overviews and AI Mode, which behave differently.

It draws on two things:

  • Our own study. We asked those four engines 210 buying questions across 21 categories and recorded every brand and every source in the 839 answers (full study and data).
  • The published research. That means the controlled experiments and the large citation studies. Where a finding is only a correlation, we say so.

TL;DR

  1. Write down the questions your buyers ask, not keywords: "best X for Y", "X that does Z", "cheaper alternative to X".
  2. Ask all four engines and record who they name and which sites they cite. They agree on the top pick only 27% of the time, so checking one engine isn't enough.
  3. Get onto the sites the answers cite. On Google's AI engines, a recommended brand's own site is cited only 5–13% of the time. The rest comes from review sites, Reddit and YouTube.
  4. Publish one focused page per question. Put the answer first, with an explicit price, a recent date, specs and an honest comparison.
  5. Cover the follow-up searches. Pages that rank for an AI Overview's sub-searches are 161% more likely to be cited 4.
  6. Make sure the page says your name next to the need. 62% of AI citations don't mention the brand that was cited 5.
  7. Re-check every week and keep pages current. A recent timestamp is one of four factors that decided citations in every model tested 1.

How an AI answer gets built

For a buying question, ChatGPT, Gemini, AI Overviews and AI Mode all do roughly the same three things:

  1. Search. The engine turns the question into one or more web searches. ChatGPT uses Bing and OpenAI's own crawl; Google's engines use Google. A page that doesn't come up in those searches can't be used.
  2. Pick sources. From the results, it picks the pages that look most on-topic, specific and current.
  3. Write. It writes a short answer and lifts passages from those pages, mostly from near the top.

So there are three gates: be found, be picked, be quoted. Better writing only helps with the third. The steps below cover all three.

Step 1: Write down the questions your buyers ask

Start with 10–20 questions a buyer would type before they've heard of you. Three kinds cover most of it:

  • Discovery: "best accounting software for freelancers"
  • Problem: "how do I stop missing follow-ups with leads"
  • Comparison: "cheaper alternatives to Zendesk"

Leave your own brand name out. "Is [your brand] good" tells you nothing about whether you get recommended.

The use case matters more than you'd think. In our study, "best running shoes for wide feet" brought up Topo Athletic and Altra, which the general running-shoe questions rarely led with. AI Mode put Topo Athletic first. For "best protein powder for women", the four engines gave four different #1s: Promix, Transparent Labs, Legion and Ritual. The specific questions are where smaller brands get in, so write the specific ones.

Step 2: Ask all four engines, and write down two things

Ask each question on ChatGPT, Gemini, Google (for the AI Overview) and Google AI Mode. For each answer, write down:

  1. Which brands it recommends, in order. Count names in headings, bold or links, not passing mentions.
  2. Which sites it cites. They're the source links in the answer.

Why all four? Because they don't agree. In our study:

  • The four engines put the same brand first on 54 of 199 questions (27%).
  • 16% of recommendations came from a single engine. For that question, the other three didn't name that brand.
  • Google's own two products shared only 53% of their recommendations for the same question. That's less than ChatGPT and Gemini, which shared 58%.

Do this by hand once and you'll learn a lot. Do it every week and you'll see what changes. That second part is what tracking tools are for.

Step 3: Get onto the sites the answers cite

This is the step most businesses skip, and for Google's engines it matters most.

Look at the sources you wrote down in step 2. In many categories, one site shows up again and again. In our study:

CategoryMost-cited siteShare of answers citing it
Small business credit cardsnerdwallet.com83%
Running shoesrunrepeat.com79%
Home security systemssafehome.org73%
Online therapyhelpguide.org70%
Travel insurancesquaremouth.com68%
Mattressessleepfoundation.org59%
AI meeting notetakersjusttalkingtech.medium.com43%

That last one is a single Medium blog. It was cited in 7 of 10 Gemini answers about AI notetakers.

Your own site plays a smaller part than you'd expect. When an engine recommended a brand, it also cited that brand's own site:

ChatGPTGeminiAI OverviewsAI Mode
57%5%13%11%

On Google's engines, a recommendation mostly comes from what other sites say about you. Google's engines also lean on Reddit and YouTube. YouTube was cited in 59% of AI Overviews answers, and Reddit in 38% of AI Overviews answers and 35% of AI Mode answers. ChatGPT cited neither in our capture.

What to do:

  • Get onto the lists that are already cited. Find the two or three pages cited most for your questions. If you belong on their list, contact the author with what's new and what's different about you, plus a way to try it. These pages already sit in the engine's trusted set.
  • Be present where real users discuss your category. That means answering Reddit threads honestly as yourself, and getting reviewed in YouTube videos. Engines read these because people trust them, so anything that looks fake works against you.
  • Keep your listings complete on the review and comparison sites in your category.

Step 4: Publish one focused page per question

The strongest controlled evidence comes from What Gets Cited (SIGIR '26). It ran 252,000 trials across six models and 18 page factors. Four factors decided whether a page was cited in every model tested. Failing any one of them could rule a page out:

  1. It matches the topic of the question.
  2. It states a price.
  3. It has a recent timestamp.
  4. It ranks high in the list of results.

Next came specs, comparisons, evidence, depth and confident rather than hedged language. Formatting by itself had no measurable effect. Headings and bullets help people read. They don't get a page picked. 1

Then there's what the engines lift once they've picked a page. Kevin Indig analysed 18,012 ChatGPT citations 2:

  • 44.2% of citations come from the first 30% of the page. Put the answer first.
  • Cited passages are almost twice as likely to use definite language: 36.2% versus 20.2%. "X is a Y for Z" gets lifted. "X might be worth considering" doesn't.
  • 78.4% of the citations tied to a question come from headings. Write the buyer's question as an H2 and answer it straight after.
  • Cited text names things. Its entity density is 20.6%, against 5–8% in normal English: brands, products, prices, numbers.

On length: in a separate study of 815,000 query-page pairs, focused pages beat "ultimate guides" for ChatGPT citations 3. One page per question, complete within that question, works better than one page that tries to cover everything.

A page that does all of this has:

  • A title that is the question, in the buyer's words.
  • A first paragraph that answers it, naming the options, including yours.
  • A price for every option. If yours is "contact us", the engine has nothing to quote.
  • A comparison table with specs and prices, and a source for each one.
  • An honest "who it's not for" section. It reads as confident and balanced, which beats promotional tone.
  • A visible "updated" date that you keep true.

Step 5: Cover the follow-up searches

Engines don't run only your exact question. They also search related sub-questions ("fan-out" queries). Surfer looked at 10,000 keywords and 33,000 of the fan-out queries behind their AI Overviews 4:

  • Pages that rank for the main query plus at least one fan-out query are 161% more likely to be cited.
  • The more fan-outs a page ranks for, the more it gets cited (Spearman ρ = 0.77). That's a correlation, not proof of cause.
  • 67.82% of pages cited in AI Overviews don't rank in Google's top 10 for the main query.

So you don't need to beat the big sites on the head term. Look at what a buyer asks next: price, setup time, "does it work with X", "vs Y". Answer those on the same page, or link to pages that do.

Step 6: Make sure the answer names you

Being cited isn't being recommended. Semrush found that 61.7% of AI citations are "ghost citations": the answer uses the page but doesn't mention the brand behind it. That was across ChatGPT, AI Overviews, Gemini and AI Mode, 115 prompts in 14 countries 5.

The passage the engine lifts has to carry your name and the need it answers, in the same sentence:

  • Weak: "Our tool helps small teams manage leads."
  • Strong: "Acme is a CRM for teams of 2–10 that costs $12 per user per month and syncs with Gmail."

Do the same on other sites. A review that says "Acme is the cheapest CRM with a Gmail sync" teaches the engine the link between you and the need. A review that says "great product!" teaches it nothing.

Step 7: Re-check every week and keep pages current

AI answers change. Our study is running for a week for that reason, and the first day's top picks already differed by engine in 16 of 21 categories. A recent timestamp is one of the four gatekeepers 1. So:

  • Re-ask your questions every week on all four engines, and note who moved in and out.
  • Update a page when its facts change (price, features, comparisons), then update the date. Don't change only the date.
  • Watch the cited sites. When a new review page starts getting cited, that's your next outreach target.

What doesn't work (or isn't proven)

  • Formatting tricks. In the SIGIR '26 experiment, content structure alone had no measurable effect 1.
  • Promotional tone. It made no consistent difference across models 1. Specific evidence did.
  • Keyword stuffing. It did no better than leaving the page alone in the original GEO experiments 6. The same paper found that adding quotations, statistics and cited sources helped. Those were measured in a fixed test setup, so treat them as direction, not a promise.
  • Rewording old pages without new facts. The engines are selecting for specifics. A rewrite with nothing new gives them nothing new to quote.
  • Checking only ChatGPT. It's one of four engines, and in our data it often disagrees with the other three.

Checklist

Done?
10–20 buyer questions written down, none naming your brand
All four engines checked; brands and cited sites recorded
The 2–3 most-cited pages per question identified, and you've contacted them
One focused page per question: answer first, prices, comparison table, date
A sentence on the page with your name, your category and who it's for
Follow-up questions (price, setup, integrations, "vs") answered
Real presence on Reddit and YouTube where your category is discussed
Weekly re-check scheduled

FAQ

How long until ChatGPT recommends me?

Nobody can honestly promise a timeline. Pages have to be crawled and indexed, and the review sites have to update their lists. We're doing this for our own product, which had 0 recommendations in our category on day 1, and we'll publish what happens.

Do I need to rank #1 on Google first?

No. Surfer found that 67.82% 4 of pages cited in AI Overviews aren't in Google's top 10 for the main query. Ranking for the follow-up questions matters a lot.

Is ChatGPT different from Google's AI?

Yes. In our data, ChatGPT linked the recommended brand's own site 57% of the time, and Google's engines 5–13%. Google's engines cite Reddit and YouTube heavily, and ChatGPT didn't cite either. You need both your own pages and third-party coverage.

Should I add an llms.txt file or schema markup?

None of the studies above tested them as a factor in being recommended, so we can't say they help. Do them if they're cheap for you, but they're not where the evidence is.

Can I just ask ChatGPT whether it recommends me?

You can, but one answer is one sample. Ask in a fresh chat, ask the questions a buyer would ask (not "do you recommend Acme?"), and check all four engines.

Sources

Our data: the AI recommendations study, November 2026. 839 answers, 21 categories and 4 engines, with the method and the data included.

  1. Vishwakarma, Kumar and Jamidar, What Gets Cited: Competitive GEO in AI Answer Engines, SIGIR '26. 252,000 trials, 6 models, 18 factors.
  2. Kevin Indig, The science of how AI pays attention, Search Engine Journal, 17 February 2026. 18,012 ChatGPT citations.
  3. Kevin Indig, Shorter, focused content wins in ChatGPT, Growth Memo, 13 April 2026. 815,000 query-page pairs.
  4. Surfer, Query fan-out impact. 10,000 keywords, 33,000 fan-out queries.
  5. Semrush, The ghost citations study, 9 June 2026. 3,981 domain appearances, 115 prompts, 14 countries.
  6. Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024.

Update log

  • 17 November 2026. First version.
  • 24 November 2026 (planned). Update the study numbers with the full week of data.

Supercite does steps 2, 3 and 7 for you. It asks your buyers' questions on ChatGPT, Gemini, AI Overviews and AI Mode every day, shows who gets recommended and which sites the answers cite, and tells you what to fix. Start a 3-day trial.

All posts →

Make AI search your next revenue channel

Start free trial