Education · AI visibility
How AI decides who to recommend.
When a customer asks an AI assistant "who should I call," the answer isn't a ranking — it's a recommendation, composed in seconds from signals most businesses have never managed. Here's how each platform builds that answer, and what actually moves it.
The shift
From ten blue links to one name.
Search built an entire industry around ranking on a page of results. AI assistants ended the page: they answer the question, and the answer typically names one to three businesses. There is no position four in a ChatGPT answer. Industry tracking projects the share of local business discovery flowing through AI assistants growing from roughly 6% to 45% — while analyses have found only about 1.2% of local businesses are recognizable to ChatGPT at all. The rest simply don't exist when the question gets asked.
- 2022ChatGPT launches — conversational AI goes mainstream in a single winter
- 2023AI assistants connect to live web search; answers start citing real businesses
- 2024Google ships AI Overviews — AI answers appear above traditional results
- 2025AI assistants become a normal way to find local services; agentic booking begins
- TodayThe recommendation layer is consolidating — and most local businesses still haven't looked at what AI says about them
The platforms
Four assistants. Four different ways of knowing you.
Want the deeper story on each — timelines, rivalries, and the standoff that reshaped government AI? Read the LLM field guide.
Training memory plus live search
ChatGPT blends what it learned in training — the businesses prominent enough to be "known" — with live web search when the question needs current, local answers. Editorial mentions, best-of lists, and a consistent web footprint decide whether you surface from either path.
Wired into Google's index
Gemini draws on Google's ecosystem — Search, Maps, and Business Profile data. Your GBP completeness, review volume and recency, categories, and photos aren't just map-pack factors anymore; they feed the AI answer directly.
Retrieval-first, citation-driven
Perplexity searches the live web for nearly every answer and cites its sources. If trustworthy pages — directories, local press, industry lists, your own well-structured site — say you're the one to call, Perplexity repeats it, with a footnote.
Careful reasoning over retrieved sources
Claude pairs strong reasoning with web search, weighing the credibility and consistency of what it finds. Conflicting phone numbers, mismatched names, and thin service pages read as uncertainty — and uncertain businesses don't get recommended.
The signals
What actually moves the answer.
Entity consistency. Same name, address, and phone everywhere. AI cross-references sources; every mismatch is a reason to name someone else.
Reviews — volume, recency, and content. AI doesn't just count stars; it reads what reviews say about the services being asked about.
Structured data. Schema markup tells machines exactly what you do, where, and for whom — retrieval-ready instead of guess-required.
Editorial presence. Best-of lists, local press, and industry directories are exactly the third-party pages retrieval-first platforms lean on. Absence from them is the most common gap we find in audits.
Pages AI can actually use. Clear service pages, real city pages, plain answers to the questions customers ask — content built for retrieval, not just keywords.
None of this is guesswork on our end: our tracking platform queries every major assistant against your market's real prompts on a schedule, so movement gets measured, not assumed.
The obvious next question
So… what does AI say about you?
The audit answers it with live queries from your market — your name in the answers, or its absence, on screen in the first meeting.
Get your AI visibility audit