A customer opens ChatGPT, types “who is nearby”, and reads the answer. That answer, not the local results page alone, is where a buying decision for a local business now forms.
The scale is no longer marginal: AI Overviews appeared for an average of about 68% of local searches in Whitespark’s Q1 2025 study. If you sell local visibility, this shift is already inside your book of clients, and it is the terrain behind Found by AI.
TL;DR
The buying moment has moved into a chatbot answer. A Google Business Profile still feeds one of the mechanisms that produce it, but it is no longer the whole of it, because each chatbot reads a business differently. Worked client by client, the shift stays invisible; the fix is to know what each chatbot reads, once, and apply it everywhere.
How agencies deliver local visibility today
Local visibility is a settled service. You keep a client’s Google Business Profile current, collect reviews, hold citations consistent across directories, and make sure the site answers the query a nearby customer types.
That offer rests on one premise: a nearby customer decides which business to call on the local results page.
What breaks when the answer comes from a chatbot
AI usage for local search jumped from 6% (2025) to 45% (2026), a 7.5x increase in a single year, per Marketing Code, 2026, recorded via Evolve Media. The customer no longer has to open a results page.
The failure that follows is quiet. A client slips out of the answer its customers now read, and nothing in the agency’s tooling says so. You find out when the client calls. No view spans the book:
- whether a client still appears in the answer;
- which chatbot stopped surfacing it, and when;
- whether the drop is one client’s accident or a pattern across the book.
Why each chatbot reads a business differently
An answer is not a ranking. Each chatbot retrieves through its own mechanism, so a business that ranks well in the local results is not read the same way once the answer is assembled. One platform is wired natively to the local data a profile feeds; others crawl the open web and weight freshness, rewarding what changed recently over what has been stable.
The mechanism decides whether your client appears, and the mechanisms do not agree with one another.
The engine I run keeps one golden rule per chatbot, each keyed to that platform’s own retrieval mechanism , and each rule activates when that chatbot’s average score is significantly below the panel median.
The cost of working client by client
The mechanism is usually learned inside one client’s problem, at that client’s expense, then left in an email thread or the head of whoever solved it, so the next account starts close to zero.
Rebuilt per account, the same knowledge is paid for again and again, and no account benefits from what the last one taught.
The objection: search is still the largest surface
The objection is fair. Search remains the largest surface for local discovery, and a Google Business Profile still earns its place. A profile is a necessary input, because one chatbot platform reads it natively.
What changed is the claim a profile carries. It is no longer the whole answer, and the shift runs one way, in the growth of AI use for local search and in the share of local queries that now carry an AI-generated answer. A profile is necessary now, and it is not sufficient, whatever the search engine’s current share.
Knowing the mechanism once, and running it across the book
The alternative is not more effort per client; it is the same knowledge held once. I keep the rules in one cascade of 22 rules across four sets, applied by decreasing priority, so a universal fix outranks a platform-specific adjustment deterministically:
- Set 1 applies to every platform.
- Set 2 addresses local results specifically.
- Set 3 addresses AI chatbots specifically.
- Set 4 holds one golden rule per chatbot.
Scoring separates four geographic dimensions instead of one local bucket: brand, town, department, and region, with the broad-local signal read as the mean of the town, department, and region layers. One rule set then serves different local footprints.
That system is mine: the rules and the scoring are one working stack I keep, and who I am is the operator behind it. Watching it run across a book of clients, rather than one account at a time, is its own instrument.
Deliverability is the rule that keeps this usable: a recommendation the non-technical owner cannot carry out fails quietly through the agency.
If your offer is local visibility and you want to judge a system against what the chatbots actually read, let’s talk.