How AI assistants decide which businesses to recommend is a naming event. AI Overviews hit 2 billion monthly users in 2025. About 1.2 percent of local businesses get named in ChatGPT. Entity plus a specific page wins.
How AI assistants decide which businesses to recommend is a naming event, not a ranking event. The model retrieves a few passages, checks whether it can verify the business as a real entity, and then says a name. Google's AI Overviews passed 2 billion monthly users in mid-2025. ChatGPT was at 800 million weekly users by October 2025. Most local firms never get named.
I study behavioral psychology, and the user-side change is small to describe. People used to scan ten links and choose. Now they ask, and a short list appears. If you are not on it, you were never in the running. This page owns the recommendation query. How engines pick sources is how ChatGPT and Perplexity decide sources to cite. How to run the checklist is how to get a small business cited by ChatGPT.
They do not run your Google rank as a score. They gather passages that look like answers, prefer pages a crawler can parse, and then need an entity they can attach a name to: a consistent NAP, a Google Business Profile, reviews, and a site that says the same facts. Search Engine Land's 2026 local-visibility coverage of SOCi's study (about 350,000 locations) put ChatGPT recommendations at roughly 1.2 percent of local businesses. Being named at all is rare. That is not a scare line. It is a market that is still empty.
SparkToro's 2024 zero-click study found 58.5 percent of US Google searches ended without a click to the open web. Digiday, citing Google's overview rollout, reported click-through falling further when an AI Overview is present (8 percent versus 15 percent without, in the figures they published). The recommendation inside the answer is the storefront.
Five signals show up again and again in what actually gets named:
Princeton's GEO paper (arXiv:2311.09735) found statistics, quotations, and citations can raise visibility in generated answers, with the lift depending on the query. That is still the cleanest public experiment. It is not a reason to invent stats. Invented stats get you dropped when a higher-authority source disagrees.
HTTP Archive's Web Almanac (2024) found JSON-LD on a minority of pages. A page with no Organization or LocalBusiness schema makes the model guess. Guessing is expensive, so it names the business whose facts already agree across maps and the official site. Structured data is the cheap half of that. The other half is entity SEO: the same name, address, phone, and category everywhere.
The second gap is query shape. Assistants recommend "a plumber in Framingham who answers after hours," not "best brand in America." Entrepreneur's 2026 AEO notes on hyperspecific service pages are directionally right: one URL per specific ask. A homepage that says "we do everything" is hard to lift into a recommendation.
The third gap is measurement. Opt-in analytics skip bots. Edge logs show GPTBot and Bingbot arriving while GA4 stays quiet. If you only watch the human dashboard, you will conclude assistants are not reading you when they already are. Server-level counts are how we saw Copilot cite a contractor site hundreds of times in a month while local peers showed nothing.
Lead with the service, the place if you have one, and a constraint you actually meet. "We host the site, write, and take the night shift so the owner does not log into a server" is recommendable. "Full-service digital excellence" is not. Keep hours, service area, and a human name on the same page the schema describes.
Writing for featured snippets is the prose habit: question heading, 20 to 40 word answer, then depth. Local SEO for multi-location businesses is the maps habit. Do not block GPTBot to "save content." You cannot be recommended by a system you locked out.
When we build this into a site at Kief Studio, recommendation readiness is a byproduct of hosting, schema, and publishing, not a bolt-on campaign. LTFI is the hire that keeps those facts current. This article stays on the decision, not the plan card.
They retrieve answer-shaped passages, then name a business they can verify as an entity: consistent NAP, schema, reviews, and a crawlable site. Rank in Google's top 10 is not required. A quotable, specific page is.
Not reliably. Ahrefs has reported that only about 12 percent of assistant-cited URLs rank in Google's top 10 for the original query. Treat recommendation and rank as separate work.
Most lack JSON-LD, consistent entity facts, and pages that answer a specific service question in the first sentences. Many also block AI crawlers. The model then names whoever is easier to verify.
Prompt the exact buyer questions in ChatGPT, Perplexity, Gemini, and AI Overviews, and log whether you are named. For Copilot, use Bing AI Performance. Do not trust opt-in analytics for crawler traffic.
No. Local recommendation rates are still low. A specific, sourced, crawlable page with matching maps data can still be the first clean answer in a category.
An AEO audit in a week: ten real questions, who gets cited, one URL per query, extractable first sentences, schema, IndexNow. This domain: 13 cited URLs of 147 posts in Bing AI Performance (29 Aug 2026).
A Bing index still matters for Copilot citations. OpenAI: ChatGPT search rewrites queries and may cite partners. This domain: 13 cited URLs of 147 posts (Bing AI, 29 Aug 2026). IndexNow with a real key file. One query, one URL.
Getting a regulated small business cited by ChatGPT: OpenAI rewrites queries and may cite partners. Put license, town, and a 50-word answer in HTML. This domain: 13 cited URLs of 147 posts (Bing AI, 29 Aug 2026).
Work With Us
Kief Studio builds, protects, automates, and supports full-stack systems for businesses up to $50M ARR.
Newsletter
Strategy, psychology, AI adoption, and the patterns that actually compound. No spam, easy to leave.
Subscribe
