Here is the short answer: every AI assistant recommends the business it can read, verify, and corroborate across the open web - but each one reads a different web. ChatGPT builds its picture of you from Bing. Claude builds it from Brave. Perplexity leans on Yelp and Tripadvisor. Gemini reads the Google stack you already know. When someone asks any of them for the best plumber, salon, or gym in your city, the assistant pulls what it can verify from its own sources and names the business it is most confident about. So the job is not to chase a ranking on one platform. It is to become the most legible, most corroborated option in your category and your city, everywhere the machines look.
The stakes moved this year. Around 37 percent of consumers now start product and service searches inside an AI tool instead of a search engine, and ChatGPT alone crossed a billion monthly users in mid-2026. The traffic these assistants send is still small - about 1 percent of visits for most local sites - but Seer Interactive measured AI-referred visitors converting at 16 percent against 1.8 percent for Google organic. Someone who arrives because a machine named you specifically is close to a buyer already. Below is the full playbook as I run it with clients: how each assistant sources its answers, what to build in which order, the tool category to skip, and how to measure whether any of it is working.
How do AI assistants decide which local businesses to recommend?
AI assistants do not rank businesses. They assemble an answer: the model takes the question, runs a live search against its index, reads the top results, and synthesizes a recommendation from whatever it can verify across those sources. Google ranks pages and shows you ten of them. An assistant reads twenty and names two. There is no position to win, only a confidence threshold to clear.
Confidence is the mechanism to internalize. When the model finds your business described the same way on your website, your profiles, a review platform, and a local news mention, each source corroborates the others and your odds of being named climb. When it finds two addresses, three phone numbers, and a website it cannot parse, it does not average the versions. It quietly reaches for a competitor it is more sure about. Every tactic in this guide is one move in the same game: raising the machine's confidence that you are the real answer.
Which search engine does each AI assistant actually use?
None of the major assistants run on Google search, except Google's own. Each one has its own backbone, which means a business can be prominent in one assistant and invisible in another, and the fix is different for each.
| Assistant | Search backbone | What it reads for local | Your first move |
|---|---|---|---|
| ChatGPT | Bing's index | Business sites, review platforms, local press in Bing's top results | Claim Bing Places, get onto pages Bing ranks |
| Claude | Brave Search | Brave's top ten organic results, near-verbatim | Check your Brave rankings today |
| Perplexity | Its own index plus data deals | Yelp, Tripadvisor, sources with clear authorship | Make your Yelp profile immaculate |
| Gemini and AI Overviews | Business Profile, Maps, the entity graph | Perfect your Google Business Profile | |
| Microsoft Copilot | Bing's index | Bing Places plus Bing organic | Inherits your ChatGPT work |
| Grok | Live web plus X | Web results and X activity | Treat as a secondary surface |
How does ChatGPT find local businesses?
ChatGPT runs a live Bing search, scans roughly the top twenty to thirty results, and picks what to cite by its own criteria - about 87 percent of its citations match Bing's top results. It does not open Bing Places directly, but Bing Places feeds the local data Bing surfaces, so claiming it is still the highest-return hour most local owners have never spent. One detail almost nobody knows: Whitespark's research found Facebook is the dominant review source inside Bing's local ecosystem, appearing on roughly one and a half times as many businesses as the next platform. The Facebook page you stopped updating in 2022 is quietly feeding the most-used AI assistant in the country.
How does Claude find local businesses?
Claude searches through Brave, an independent index - Anthropic lists Brave as a subprocessor, and analysis found 86.7 percent of Claude's citations match Brave's top ten organic results. Claude barely re-ranks: if you are not in Brave's top ten for a query, you effectively do not exist for it. This matters more than it used to, because Claude's share of AI referral traffic in business categories jumped from around 1.4 percent to 18.5 percent in a year. The check takes two minutes: search your category and city at search.brave.com and see who actually appears. Brave crawls the open web independently, so the same server-rendered, schema-marked pages that help elsewhere are what get you in.
How does Perplexity find local businesses?
Perplexity cites more sources per answer than any other assistant - around eight, roughly three times ChatGPT - and for local queries it leans hard on structured review platforms. Yelp holds about 62 percent of Perplexity's local-citation share, nearly five times the next source, and a January 2025 partnership wired Tripadvisor's billion-plus reviews directly into its answers. If your Yelp profile is a ghost town with a wrong phone number, Perplexity is reading that ghost town out loud to your prospects.
How do Gemini and Google AI Overviews pick businesses?
Gemini and AI Overviews read the Google stack: your Business Profile, Maps, reviews, and the entity graph, which is why the local SEO you have already done transfers here most directly. The scale is the story - AI Overviews now appear on roughly 48 percent of all queries and over 80 percent of local service queries, and when one appears, organic click-through drops by about 61 percent. Notably, around 60 percent of AI Overview citations for local searches point to third-party pages like directories, publishers, and forums rather than business websites. Being named on pages Google trusts now beats ranking your own page beneath an answer nobody scrolls past.
Which business profiles do the AI models actually read?
Three profiles carry most of the structured-data weight: Google Business Profile, Bing Places for Business, and Apple Business Connect. Structured profiles are the easiest thing for a machine to trust, which makes them the foundation everything else corroborates.
- Google Business Profile, done to the AI standard: primary category set to your highest-revenue service, every relevant secondary category added, service descriptions written in full plain-language sentences, service area, hours, photos, and Q&A all current. The models read the categories and descriptions, not your intentions.
- Bing Places for Business, because it feeds both ChatGPT's and Copilot's view of your market, and most of your competitors have never claimed it. That is the opening.
- Apple Business Connect, which feeds Siri and Apple Maps. Quick to set up, rarely done.
- Name, address, and phone identical everywhere, down to the "St" versus "Street". A machine reading four slightly different addresses does not average them. It loses confidence.
- The same one-sentence business description reused across all three profiles, your site footer, and every directory that lists you.
Which reviews and mentions actually move AI recommendations?
The signals outside your own website split into two layers: reviews, which teach the models what you do, and editorial mentions, which prove you are who you claim. Most owners over-invest in star averages and under-invest in both of these.
What should customers actually say in reviews?
The models care more about what reviews say than what they score. A detailed review that names the service, the outcome, and the neighborhood hands the model language it can repeat: "they re-piped a rowhome in Manayunk in two days and the quote held" is worth more than ten reviews that say "great service." So the ask changes from "leave us five stars" to "mention what we did and where." Then route the asks by platform, because each assistant reads a different review economy: Google reviews feed Gemini and get pulled into ChatGPT answers, Facebook reviews dominate Bing's local layer, Yelp owns Perplexity, and Tripadvisor matters if you touch travel or food. Keep them recent and reply to every one - recency and responses are both text that confirms a live, attentive business.
How do I get named in best-of lists?
Editorial mentions are the differentiation layer. When BrightLocal reverse-engineered 800 ChatGPT local searches, about 58 percent of citations were businesses' own websites, 27 percent were editorial - articles, blog posts, best-of lists - and only 15 percent were directories. Whitespark's 2026 ranking-factors survey backs it up: three of the top five AI visibility factors are citation-related, led by presence on expert best-of lists. So the outreach list writes itself: the "best [your service] in [your city]" articles, local press roundups, your chamber and industry association pages, and the curated directories the studies actually surfaced - Three Best Rated, Expertise, Tripadvisor. One mention is a claim. Five consistent mentions across sites the model trusts is a fact. Corroboration is the currency.
Does Reddit matter for a local business?
More than you would guess. Peec AI's analysis of 30 million cited sources found Reddit is the single most-cited domain across AI engines, ahead of YouTube, because the models treat threads as candid human judgment. For a local business that means the "who do you recommend" threads in your city subreddit are being read aloud by machines for years after they are posted. Participate as yourself, disclose you are the owner, answer the question better than anyone else, and never astroturf - fake accounts get caught and archived forever. One caution before you build a strategy on any single platform: when a parameter changed in late 2025, ChatGPT's share of Reddit citations swung from roughly 60 percent to 10 percent in six weeks. Sources rotate. Corroboration across many surfaces is the only durable position.
How do I make my website readable to AI crawlers?
Server-rendered HTML is the floor. Many modern templates render their content with client-side JavaScript, which means the words a human sees are not in the raw HTML a crawler first receives - and the model may never read your services or your city at all. View your page source and check that your core answer is actually in it. On top of that floor, two things do the work:
- Structured data. Add LocalBusiness or ProfessionalService schema, Service schema for each offer, Organization schema with sameAs links pointing at your profiles, and FAQPage schema on any page with a real question-and-answer section. Schema hands the model a labeled version of the facts instead of making it guess.
- Extractable writing. Open every page with a direct answer, use question-shaped headings, numbered steps for processes, a comparison table for tradeoffs, and a visible FAQ in real HTML. Write the way you want to be quoted, because being quoted is the entire game.
- Freshness. Multiple 2026 studies found the assistants cite stale content less each quarter. A page updated with current numbers beats the same page from 2024, so revisit your money pages on a schedule.
Do I need an llms.txt file?
No. Almost every AI-visibility checklist published this year tells you to add one, and the data says skip it: Ahrefs studied 137,210 domains in May 2026, found 28 percent had published the file, and 97 percent of those received zero AI traffic from it. The bots do not look for it. Put the hour into schema and business-information consistency, where the models are actually reading.
Can a tool put my AI rankings on autopilot?
No. Nobody controls how these models answer - including, in any precise sense, the companies that build them. Searches for AI visibility tools are up roughly 1,900 percent year over year, and a new "AEO on autopilot" product launches weekly, so it is worth being exact about what these tools can and cannot do. The legitimate ones are measurement layers: they run prompts at scale and report how often you are named. That is useful. But none of them write your content, fix your schema, claim your Bing profile, earn a best-of mention, or generate a detailed review - the inputs the models actually weigh. A dashboard that watches the scoreboard does not play the game.
And the guarantee version of the pitch is the old "guaranteed page one on Google" scam with a new coat of paint - I wrote about why you should run from guaranteed rankings, and every word applies double here. The Reddit citation swing above is the proof: a single upstream parameter change moved a source from 60 percent to 10 percent in six weeks. Anyone promising you a stable, guaranteed position inside a system that volatile is describing a product that cannot exist.
You cannot buy a ranking inside a machine you do not control. You can only become the answer it keeps finding.
How do I measure whether AI search sends me customers?
Three layers, from complimentary to paid. First, GA4 now does part of this natively: as of 2026 it classifies recognized assistants into their own channel, visible under Reports, then Acquisition, then Traffic acquisition. But between a third and two thirds of AI visits arrive with no referrer and hide inside Direct, so treat the reported number as a floor. For older properties or finer control, build a custom channel group (Admin, then Data display, then Channel groups) with a source-matching condition:
Channel: AI Assistants
Condition: Session source matches regex
chatgpt.com|chat.openai.com|perplexity.ai|claude.ai
|gemini.google.com|copilot.microsoft.com|grok.com
|meta.ai|deepseek.com|you.comSecond, run a monthly prompt audit by hand: from an incognito session, ask each assistant the five questions your customers would actually ask, note who gets named, and keep the answers in a spreadsheet. It costs nothing and tells you who to study. Third, when you want a continuous read, the trackers tier cleanly in 2026: LLMrefs runs a single keyword at no cost and 79 dollars a month for 500 prompts, Peec starts around 75 euros a month, and Profound starts at 499 dollars self-serve with enterprise deals in the tens of thousands. For most local businesses the manual audit plus GA4 is plenty for the first six months.
What is the 90-day plan to show up in AI search?
The order matters: profiles first because they are fast and foundational, then the site, then the slower corroboration work that compounds.
- Days 1 to 7: claim and complete the profile layer. Google Business Profile to the AI standard, Bing Places, Apple Business Connect, and one NAP sweep so every listing matches to the character.
- Days 8 to 21: fix the site floor. Confirm your content is server-rendered, install LocalBusiness, Service, Organization, and FAQPage schema, and rewrite your money pages answer-first.
- Days 22 to 45: start the review engine. Change the ask to "mention what we did and where," reply to everything, and build coverage on the platform your assistants read - Facebook and Google at minimum, Yelp if Perplexity matters to your customers.
- Days 46 to 75: earn mentions. Pitch the best-of lists in your city, get onto your chamber and association pages, answer the standing Reddit threads as yourself, and claim the curated directories the studies surfaced.
- Days 76 to 90: baseline the measurement. Set up the GA4 channel, run your first incognito prompt audit across ChatGPT, Claude, Perplexity, and Gemini, and record who gets named.
- Ongoing: publish one page a month that answers a real customer question with current numbers, because fresh, specific content is what all four assistants keep choosing to quote.
Notice that none of this is a clever trick. A local business is rarely invisible to AI because it missed one tactic. It is invisible because of fragmentation: an address that disagrees across a dozen tools, a website the models cannot read, reviews scattered and unanswered, no schema, a Bing profile nobody claimed. The machines meet a business that cannot describe itself consistently, and they move on.
That same fragmentation is why your software bill is too high - every disconnected tool holds its own slightly different version of the truth, and you pay for all of them to disagree. The two problems are the same problem, and that is the work I do: one audit that finds where you are invisible to AI and where you are overpaying for tools that overlap, then a single system that fixes both, so the savings help fund the visibility work. Fire a standard agency and you lose the tactics the day you stop paying. Work with me and the infrastructure stays installed. That is by design.
Common questions
Does ChatGPT use Google to find local businesses?
No. ChatGPT's live search runs on Bing's index, though it has been observed pulling Google review data into answers. Claiming Bing Places and earning placement on pages Bing ranks are the levers that reach it.
Which AI assistant should I optimize for first?
ChatGPT, on volume - it still carries roughly two thirds of AI referral traffic. But about 80 percent of the work overlaps: consistent profiles, readable site, detailed reviews, and editorial mentions raise your visibility in all four assistants at once, so do the fundamentals once rather than optimizing platform by platform.
Do I need an llms.txt file?
No. Ahrefs' May 2026 study of 137,210 domains found the assistants do not read it. Spend the time on schema, server-rendered content, and consistent business information instead.
Can an agency guarantee my business will rank in ChatGPT or Gemini?
No, and a guarantee is the tell to walk away. Nobody controls model output, sources rotate without warning, and the honest version of this work is fundamentals plus measurement, not promised placement.
How long before AI assistants start recommending my business?
Plan in weeks to months, not days. The assistants have to recrawl your updated profiles and site, then see you corroborated across enough independent sources to clear their confidence bar. Recent reviews and fresh content shorten the cycle.
