AI Business Guides

How to Get Your Business Recommended by ChatGPT

By Scott McIntosh, Founder of DigitalTreehouse · Updated November 2026

Illustration of a small shop lit by a beam of light, for the article How to Get Your Business Recommended by ChatGPT

Open ChatGPT and type the question your best customer would ask right before hiring you. "Best commercial HVAC contractor in Williamson County." "Estate planning attorney in Franklin who handles farmland." "Chiropractor in Brentwood that takes walk-ins." Read the answer.

If your business is in it, this article will help you stay there. If it is not, this article explains exactly why, and what to do about it, using the research on how ChatGPT actually chooses what to cite rather than guesses.

Why this matters now

In August 2026, ChatGPT passed 1 billion weekly active users, up from 900 million in February. A growing share of those sessions are recommendation questions: who to hire, where to go, what to buy. When an assistant answers, it typically names three to seven options, not ten blue links. Being in that short list is a different game from ranking on page one of Google, and it is a game most small businesses have not started playing.

The traffic that does arrive from AI assistants is unusually valuable. Adobe Analytics measured a 16 percent conversion uplift for AI-referred visitors versus non-AI sources in U.S. retail, and Similarweb has found that ChatGPT accounts for roughly 87 percent of all AI referral traffic. A visitor who arrives after an AI assistant already vouched for you is closer to a decision than a visitor who clicked a search result.

How ChatGPT actually decides who to recommend

Three mechanisms determine whether you show up. Each one is a lever.

1. What the model already "knows" (training data)

ChatGPT's underlying models are trained on a large snapshot of the web. If your business has been written about, reviewed, listed, and discussed in enough places over enough time, the model has an internal sense that you exist and what you do. This is slow to build and slow to change, but it is why long-established businesses with a real footprint tend to appear even without doing anything deliberate.

2. What it retrieves at answer time (search)

For many recommendation queries, ChatGPT runs a live web search and reads a handful of pages before answering. Not every time: Ahrefs found that only 14.4 percent of 66.6 million free, logged-out ChatGPT responses in August and September 2026 involved a web search. But "who should I hire" questions are exactly the kind that trigger one, and this is where you can move the needle in weeks rather than years. Two facts about this retrieval matter enormously:

It mostly uses OpenAI's own index now, not Bing. For years the standard advice was that ChatGPT search ran on Bing. That is out of date. OpenAI has never said it uses Bing, it runs its own search crawler (OAI-SearchBot), and its head of ChatGPT testified in 2025 that the company was building its own search index. Measurements in July and August 2026 found most ChatGPT results came from OpenAI's in-house index, with the rest from Google results, and Bing's share small, possibly near zero. Two practical takeaways: make sure OAI-SearchBot can reach every page, and keep your Google visibility strong. Bing Webmaster Tools is still worth keeping, but for a different reason: since 2026 its AI Performance report shows which AI queries retrieved your content and your citation share, data no other tool gives you.

It reads far more than it cites. Analyses published in early 2026 found that only about 15 percent of the pages ChatGPT retrieves end up cited in the final answer, and that citations concentrate in the first 30 percent of a page. Getting retrieved is necessary. Getting cited requires that the retrieved page answers the question clearly, early, and specifically.

3. Which sources it trusts (citation patterns)

Several large studies have measured what ChatGPT cites. The biggest story of 2026 is how fast the answer changed.

The lesson is not "get on Reddit," and it never was. The lesson in late 2026 is that ChatGPT rewards the original source of a fact. That is good news for a small business: you are the original source for your own prices, process, service area, and results. Publish them clearly on your own site, and back them up with review platforms, directories, local press, and professional profiles that say the same thing.

The research on what makes content get cited

In 2024, researchers from Princeton, Georgia Tech, the Allen Institute for AI, and IIT Delhi published the first controlled study of what they named generative engine optimization. They tested nine content changes across 10,000 queries and measured how often the modified pages were cited in AI-generated answers.

The findings, in order of effect:

  • Adding quotations from relevant sources produced roughly a 40 percent visibility gain.
  • Adding statistics produced gains in the 30 to 40 percent range and was the single most effective strategy for many query types.
  • Citing sources produced roughly a 28 percent gain.
  • Keyword stuffing made visibility worse, by about 10 percent.
  • Smaller, lower-ranked sites saw the largest gains from these changes.

That last point is the encouraging one for a local business. The techniques that work for AI visibility are the ones a small, specific, honest business can do better than a large generic one: say precise things, cite where they come from, and quote real people.

The eight-step playbook

Step 1: Measure your baseline with a prompt panel

Before changing anything, build a list of 15 to 25 prompts your customers would actually type. Mix them across the buying journey:

  • Category: "best [service] in [city]"
  • Problem: "who can help me with [specific problem] near [area]"
  • Comparison: "[your business] vs [competitor]"
  • Attribute: "[service] in [city] that [specific feature: emergency hours, financing, specialty]"
  • Brand: "what is [your business] known for"

Run each prompt in ChatGPT (with search on), Claude, and Google AI Mode. Record whether you appear, in what position, what is said about you, and which sources are cited. Repeat monthly. This is your scoreboard.

Here is a scoring sheet we use for client audits. Score each prompt 0 to 3.

Score Meaning
0 Not mentioned
1 Mentioned in passing, no detail or wrong detail
2 Recommended with accurate description
3 Recommended first or specifically, with a reason and a citation

Sum across 20 prompts for a score out of 60. Most businesses we audit start between 4 and 15. A well-executed program moves that into the 30s within a quarter.

Step 2: Make sure you can be read at all

This is the unglamorous step that fixes the most cases of "we don't show up."

  • Check robots.txt. OpenAI crawls with user agents including GPTBot (training) and OAI-SearchBot (live search). Many security plugins and CDN defaults block "AI bots" wholesale. If OAI-SearchBot is blocked, you cannot be retrieved at answer time. Decide deliberately.
  • Keep Google visibility strong with Search Console and a clean sitemap, since much of ChatGPT's retrieval draws on Google results. Also verify in Bing Webmaster Tools, not because ChatGPT relies on Bing, but because its AI Performance report shows which AI queries retrieved your pages.
  • Render without JavaScript. AI crawlers are less patient than Google. If your key content only appears after scripts run, it may not be read. Pure-code sites (static HTML, Astro, and similar) have a real advantage here.
  • Page speed and clean HTML. Same reason.
  • Put the answer in the first 30 percent of the page. If your services page opens with 400 words about your founding story, the crawler may never reach the part that answers the question.

Step 3: Establish your entity

AI systems need to be confident about who you are. Consistency across the web is how they get there.

  • Identical name, address, phone, and description everywhere: your site, Google Business Profile, Bing Places, Apple Maps, Yelp, industry directories, LinkedIn, Facebook.
  • Schema.org structured data on your site: LocalBusiness (or the specific subtype), Service, FAQPage, Review, and Person for your key people. This is machine-readable confirmation of what you do, where, and for whom.
  • A clear "what we do, for whom, where" statement in plain text on your homepage and about page. Not a slogan. A sentence a model could quote.
  • Wikipedia only if you genuinely qualify. Wikipedia's notability rules exist for a reason; do not attempt to game them. But if your business has real press coverage, a properly sourced article is disproportionately valuable in ChatGPT specifically.

Step 4: Build the third-party footprint

This is where most of the ChatGPT-specific work lives, and where a marketing agency that only touches your website will leave you invisible.

Directories and review platforms. For recommendation queries, the studies consistently show Yelp, G2, Capterra, Avvo, Healthgrades, Houzz, Angi, and the equivalents for your industry among the cited sources. Claim every relevant profile, complete it fully, and keep it current. This is the closest thing to a direct lever on "best X in Y" queries.

Reviews, with specifics. A review that says "Great service!" is nearly useless to a model. A review that says "They replaced our 20-year-old unit in one day and the financing was 0 percent for 18 months" contains exactly the attributes a model matches against a question. Ask customers for specifics. Respond to every review with more specifics.

Community presence. ChatGPT cites Reddit far less than it did before August 2026, but it still cites it, other AI assistants such as Perplexity still lean on it, and local subreddits and industry forums are where "who do you recommend for X" threads live. Participate honestly, as yourself, answering questions. Never astroturf; models are trained on the patterns of fake recommendations too.

Press and expert citations. Being quoted in local and trade press does two things: it creates a citable third-party source, and it establishes your people as named experts. A repeatable way to do this is to monitor daily journalist-request services and answer the requests that match your expertise. We run this as a machine for clients; done by hand, it takes about 20 minutes a day.

Professional profiles. LinkedIn company pages and personal profiles are among the top-five cited domains in multiple studies. Complete them with the same specific language.

Step 5: Publish content that answers questions the way the research says

Apply the Princeton findings directly.

  • One question per page, answered in the first paragraph. "How much does a commercial roof replacement cost in Middle Tennessee?" gets a page with a specific range and the factors that move it, up top.
  • Statistics with sources. Your own data is the strongest kind: average job time, average cost, number of projects, response time. Cite external data where you use it.
  • Quotations from named people. Your technicians, your customers (with permission), industry experts.
  • FAQ sections marked up with FAQPage schema.
  • Comparison and "vs" pages that are fair. Models prefer even-handed content and will cite it for comparison queries.
  • No keyword stuffing. The study measured it. It hurts.

Step 6: Decide what to do about llms.txt

llms.txt is a proposed convention: a plain-text file at your site root that summarizes your business and points to your most important pages, written for AI systems. As of this writing, no major AI provider has confirmed it uses the file for retrieval. It costs 20 minutes, cannot hurt, and may help. Add it, keep it honest, and do not expect it to do the work of Steps 2 through 5.

Step 7: If you sell products, register with the merchant programs

OpenAI runs a merchants program for product businesses, and ChatGPT's shopping features surface products from structured feeds. If you sell physical or digital goods, getting your catalog into those feeds with complete, accurate attributes is a direct path into product recommendation answers. Service businesses can skip this step.

Step 8: Monitor monthly and correct errors

Re-run your prompt panel every month. Track the score. When ChatGPT says something wrong about you (wrong service area, discontinued service, wrong hours), trace the citation. Nine times out of ten it is a stale directory listing or an old page you forgot about. Fix the source and the answer follows within weeks.

A realistic timeline

Month Work Expected result
1 Baseline panel, crawlability fixes, Search Console and Bing verification, schema, entity cleanup Errors corrected; first appearances on brand and attribute prompts
2 Directory and review program launched, five to ten question pages published, press-request monitoring live Appearances on problem and attribute prompts; first third-party citations
3 Expand content, first press mentions land, community participation steady Appearances on category prompts in your immediate area
4 to 6 Compound: more reviews, more citations, more pages Consistent top-three recommendation on core prompts locally

Progress is real but not instant, and it is not permanent. Citation patterns moved 50 points in six weeks in late 2025, and Reddit's share fell by more than 80 percent in a single week in August 2026. Whatever you build has to be maintained.

Five things to do this week

If the full playbook feels large, these five actions take an afternoon and move the score.

  1. Run ten prompts in ChatGPT with search on and write down the result. You cannot fix what you have not measured.
  2. Check robots.txt for OAI-SearchBot and GPTBot. Unblock the search crawler if you want to be found.
  3. Search for your core service and city in Google. If you do not show up there, you will struggle in ChatGPT too, since much of its retrieval draws on Google results.
  4. Fix the top three inconsistencies in your name, address, phone, and description across Google, Bing, Yelp, and LinkedIn.
  5. Rewrite the first paragraph of your main service page so it states, in one plain sentence, what you do, for whom, and where, followed by one specific number about your work.

None of these require a budget. All of them are things competitors in most local markets have not done.

What not to do

  • Do not block AI crawlers by default and then wonder why you are invisible. Decide.
  • Do not buy "AI citation" packages that promise placement. There is no paid placement inside ChatGPT's organic answers.
  • Do not fake reviews or Reddit threads. Beyond the ethics, models are trained on the fingerprints of fake content.
  • Do not stuff keywords. The one technique proven to reduce visibility.
  • Do not treat this as separate from SEO. The same crawlability, structure, and authority signals feed both. The difference is that AI visibility weights third-party corroboration and specific, citable content more heavily.

How DigitalTreehouse approaches this

We run an AI Visibility Scan that produces the prompt-panel score above, a citation map (who is being cited instead of you, and why), and a crawlability report. For clients who want it done, our Optimized Website System rebuilds the site as clean, crawlable code, handles schema and entity work, runs the directory and review program, and operates a press-request scanner that turns journalist queries into citations every month.

The businesses that win this are not the biggest. They are the most specific, the most consistently described across the web, and the most willing to publish precise, sourced answers to the questions their customers actually ask. That is a small-business advantage, if you use it.

Frequently asked questions

Can I pay to be recommended by ChatGPT?

Not inside the organic answers. There is no placement product for the recommendations ChatGPT gives when someone asks who to hire. OpenAI has advertising on some consumer tiers and a merchant program for products, but the recommendation you are trying to earn comes from retrieval and trust signals, not spend. Anyone selling guaranteed placement is selling something else.

Does this replace SEO?

No. The same foundation feeds both: a crawlable site, clear structure, consistent entity data, authoritative third-party mentions. Two things differ in emphasis. AI visibility rewards being the original source of specific, checkable facts, and it weights third-party corroboration (review sites, directories, press) more heavily than classic SEO does. Treat it as an extension of SEO with a different scoreboard, not a replacement.

How long before I show up?

Errors and brand-query gaps usually correct within weeks once the underlying source is fixed. Attribute and problem queries follow in the second month as content and directory work lands. Competitive category queries ("best X in city") take a quarter or more and depend on how much third-party presence competitors already have. Progress is measurable monthly with the prompt panel.

Should I block AI crawlers to protect my content?

That is a legitimate choice for publishers whose product is the content itself. For a business that wants to be found and recommended, blocking the search crawler means removing yourself from the answer. Note that training crawlers and live-search crawlers are different user agents; you can allow one and block the other. Make the decision deliberately rather than inheriting it from a plugin default.

Does the answer change from person to person?

Yes. ChatGPT personalizes with memory and location, and answers vary between sessions. That is why the prompt panel uses many prompts and is run repeatedly rather than relying on a single check. Track the trend, not any one answer.

What about Claude, Gemini, Perplexity, and Google AI Overviews?

Each retrieves and cites differently. The studies show Perplexity leans on Reddit and primary sources, Google AI Overviews on YouTube and its own index, and Claude on long-form editorial content. The playbook in this article, crawlability, entity clarity, third-party footprint, and specific sourced content, moves all of them. Run the prompt panel across all of them and you will see which one needs the most work.

Is llms.txt required?

No provider has confirmed using it for retrieval. It costs almost nothing to add, so add it, but put the real effort into the steps that are proven: crawlability, entity consistency, third-party presence, and content with statistics, quotes, and citations.


Sources

  • Stocktwits via Yahoo Tech, ChatGPT tops 1B weekly users (August 6, 2026): tech.yahoo.com
  • Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024 (Princeton, Georgia Tech, Allen Institute for AI, IIT Delhi): arxiv.org/abs/2311.09735 and Princeton listing
  • Search Engine Land, AI search engines cite Reddit, YouTube, and LinkedIn most (30 million source analysis): searchengineland.com
  • 5W, AI Platform Citation Source Index 2026: 5wpr.com
  • Search Engine Journal, ChatGPT's most-cited sites shift to primary sources (Ahrefs data, October 12, 2026): searchenginejournal.com
  • Axios, Reddit fades from ChatGPT citations (August 20, 2026): axios.com
  • Inc., ChatGPT Suddenly Cut Reddit Citations by 81 Percent (August 20, 2026): inc.com
  • Falia, What search engine does ChatGPT use? No longer Bing: falia.co
  • Adobe Analytics and Similarweb AI referral figures, as compiled at aristral.com
  • OpenAI, ChatGPT for Merchants: chatgpt.com/merchants

Keep reading