The search that doesn't look like search
In BrightLocal's 2026 consumer survey, 45 percent of US consumers said they had asked an AI tool for local business recommendations — up from 6 percent a year earlier. ChatGPT led at 31 percent, with Google's AI Mode at 23 percent. That is no longer an early-adopter story. ChatGPT alone reported 900 million weekly users in February 2026.
The same survey has the number that should shape your plan: 63 percent of those users trust the recommendations, but 97 percent still verify them against reviews at least sometimes. The assistant opens the shortlist; your reviews close it. Everything below follows from that order.
Google is running the same playbook inside classic search. Semrush measured AI Overviews on 15.7 percent of queries in November 2025 across a 10-million-keyword panel, and BrightEdge's commercial-keyword tracking has them on roughly half of the queries it follows, with healthcare at 88 percent. Even the customer who never opens ChatGPT is increasingly reading an AI answer before your link.
Where the assistants get their answers
When ChatGPT or Perplexity answers "best med spa near Main Street," it is not reciting training data. It runs a live search, reads a handful of pages, and quotes them. So the practical question is which pages it reads.
For local queries the answer is mostly review platforms and directories. A Foundation Marketing study of 28 million local queries across ChatGPT, Gemini, Perplexity and Google AI Mode counted 512,680 citations for Yelp in a single quarter — 3.4 times the next source and more than its five tracked competitors combined. Over half of Yelp's citations came from "near me" queries. Semrush's three-month, 100-million-citation study shows the same pattern with different weights per assistant — and shows the weights move: ChatGPT cited Reddit in about 60 percent of responses last summer, then OpenAI adjusted something in September and it fell to about 10.
We measure this on our own sites, and our single-site numbers agree with the big studies. Two findings from our log: almost every citation an assistant took from us was lifted from a short, self-contained passage — about 200 characters that answer one question — and the single biggest mover for brand queries wasn't on our site at all. Getting our reviews onto a directory the assistants already trusted took the rate at which AI answers named us from roughly 1 percent to over half of test runs. One site's data, honestly labeled. But it matches what the citation studies keep finding: assistants trust the places that hold reviews more than they trust you.
The five levers, in the order we'd pull them
1. Reviews where the assistants already look. Google and Yelp reviews are the raw material of local AI answers, and the 97 percent verification habit means they get read twice — once by the machine, once by the human checking it. Volume, recency and answered reviews all show. If review collection isn't running every week, start there; here is the system we run.
2. A site the crawlers can actually read. Vercel and MERJ measured the major AI crawlers and found none of them execute JavaScript — they fetch your scripts and never run them. If your prices and services only appear after JavaScript runs, ChatGPT reads an empty page. Server-rendered HTML is the fix, and it's a build decision, not a marketing one.
3. Let the right bots in. Each assistant documents its crawlers: OpenAI's OAI-SearchBot surfaces sites in ChatGPT search, and a site that blocks it "will not be shown in ChatGPT search answers" — their words. Perplexity documents PerplexityBot, Anthropic documents ClaudeBot and Claude-SearchBot. Read your robots.txt once. Blocking training bots while allowing search bots is a legitimate middle position; blocking everything is invisibility you chose.
4. Answer-shaped pages. Because citations get lifted from short passages, a page that states the answer plainly — the price, the timeline, the yes-or-no — in the first two sentences of a section gets quoted, and a page that warms up for four paragraphs doesn't. Write the number, then explain it. Our prices are public partly for this reason: an assistant can't quote a "contact us."
5. One consistent identity everywhere. Assistants resolve businesses as entities. When our website and our LinkedIn page described the company differently, AI answers mixed us up with unrelated companies that shared a word of the name; making every profile tell one story fixed it. Same name, address, phone and description on the site, Google Business Profile, LinkedIn and directories — and keep the Business Profile current, which is also Google's own stated best practice for its AI features.
What each assistant reads
| Crawler to allow | Runs your JavaScript? | Where that's documented | |
|---|---|---|---|
| ChatGPT search | OAI-SearchBot (GPTBot is training only) | No | OpenAI bot docs |
| Perplexity | PerplexityBot | No | Perplexity crawler docs |
| Claude | Claude-SearchBot, Claude-User | No | Anthropic help center |
| Google AI Overviews | Googlebot (normal indexing) | Yes, via Google's renderer | Google Search Central |
What we'd skip
An honest word about llms.txt, the proposed machine-readable site guide, because "AI SEO" vendors sell it hard. Google's John Mueller compared it to the keywords meta tag, and no major assistant has said its crawlers read the file. We publish one anyway — it costs an hour and it can only help an agent that chooses to look — but we'd never bill a client for it as the strategy. Google is explicit that no special files or markup are required for AI Overviews at all.
The same skepticism applies to anything promising to "inject" your business into AI answers. The citation studies above show what actually gets read: review platforms, directories, and plain pages that answer questions. There is no side door. And AI answers are not stable — the Reddit swing from 60 percent to 10 shows the ground can move in a month, which is why we measure monthly and treat any single answer as weather, not climate.
Sources
- BrightLocal, Local Consumer Review Survey 2026: AI and trust, March 2026
- TechCrunch, ChatGPT reaches 900M weekly active users, February 2026
- Foundation Marketing, Yelp's dominance in AI local discovery, July 2026
- Semrush, The most-cited domains in AI, November 2025, and AI Overviews study, December 2025
- Search Engine Journal, BrightEdge AI Overviews data, March 2026, and John Mueller on llms.txt, April 2025
- Vercel × MERJ, The rise of the AI crawler, December 2024
- OpenAI bot documentation; Perplexity crawler documentation; Anthropic crawler documentation; Google, AI features and your website
Find out what the assistants say about you now
Ask Sunday in the chat for a free visibility check: whether ChatGPT, Gemini and Perplexity name your business, how your Google Business Profile compares against local competitors, and the top three fixes — back in two business days, no call required. For the wider picture, what an AI marketing agency actually is and the Business Profile optimization guide are the two posts we'd read next, and our AI SEO service covers this work end to end.