"Which tradesperson near me does reliable work?" — in 2026 that question no longer goes only to Google but increasingly to ChatGPT, Gemini or Perplexity. The answer is then not ten blue links but three to five names. If you are on that short list, you are in the running. If you are missing, you simply do not exist for that prospect.
That changes the rules for local businesses fundamentally — but not in the way many assume. No new secret technique is needed. What is needed is the old homework, done more consistently than before: complete profile data, contact details without contradictions, current reviews, and a website that makes facts machine-readable. Here is which levers work measurably — and which are overrated.
1. The starting position: local search now runs on two tracks
For years local visibility had one clear address: the Google local pack, the map box with three entries above the organic results. If you were in it, you got calls and directions. That mechanism still works — in 2026 it has simply gained company.
In parallel, language models produce their own local recommendations. They do not do so out of nothing: AI platforms synthesise their answers from the same sources that feed classic local search — business profiles, directories, review portals and websites. The decisive difference lies in the selection. A results list has room for twenty businesses; an AI answer has room for four.
2. The figures: how big the gap between Google and AI is
Two current surveys paint a clear picture. The SOCi Local Visibility Index 2026 evaluated over 350,000 locations across 2,751 brands. The result: the locations studied appear in the Google 3-pack in 35.9 % of cases — but are recommended by ChatGPT in only 1.2 %. SOCi puts the selectivity of AI platforms at three to thirty times that of classic local search.
At the same time demand is growing fast. The BrightLocal Local Consumer Review Survey 2026 (1,002 adults surveyed in the USA) shows the share of consumers using AI for local recommendations rose within a year from 6 % to 45 %. That makes AI the third most important discovery channel — behind Google and Facebook, but ahead of Yelp and Tripadvisor. Google's share of local business search fell over the same period from 83 % to 71 %.
A third figure explains why data maintenance carries so much weight: according to SOCi, contact details in ChatGPT answers are correct only 68.3 % of the time, and 68.0 % at Perplexity. So roughly every third local AI recommendation contains a wrong address, phone number or opening time — a direct loss of revenue for the business concerned.
The BrightLocal consumer data comes from a US sample (1,002 respondents) and does not transfer one to one to other markets. The direction of travel, however, matches what we see in client accounts. As of August 2026.
3. Google Business Profile: the foundation — and what changed in 2026
The Google Business Profile (formerly Google My Business) remains the single most important dataset a local business has. It feeds Google Maps and the local pack — and it is at the same time one of the best-maintained public sources about local companies anywhere. That is exactly why it shows up in the training and retrieval data of AI systems.
Three developments upgraded the profile in 2026:
- Ask Maps. The Gemini-powered feature has, since March 2026, answered user questions directly from profile details, website content and review text. What is not in the profile cannot be answered.
- Recency over volume. Since the core update in March 2026 Google weights the freshness of reviews more heavily than their sheer number. Active profiles with a moderate review count gain; stagnating ones with a large old stock lose.
- Video verification. Since early July 2026 Google has offered eligible profiles verification by video as an alternative to the postcard — considerably shortening the route to a verified profile.
Concretely that means: set categories completely and precisely (primary category plus fitting secondary ones), maintain services and products individually, keep opening hours including holidays current, add photos at intervals, and work the question-and-answer section actively. It is unspectacular — and precisely for that reason the area where most businesses give visibility away.
4. NAP consistency: the underrated lever
NAP stands for name, address, phone. The term comes from classic local SEO but has gained new weight under AI conditions. A language model has to infer from scattered sources that "Miller Plumbing Ltd" on the High Street and "Plumbing Miller" on High Str. are the same business. The more contradictory the data, the less certain the assignment — and uncertain entities get recommended less often.
Where discrepancies typically arise
The most common causes are harmless and effective all the same: a move that was only carried through in the Google profile; an old landline number still sitting in three directories; a legal-form suffix that is sometimes carried and sometimes not; a second phone number from an old advertising campaign. Each of these is unproblematic for humans — for a machine it is a contradiction.
How to create consistency
Fix one binding spelling — exactly as it appears in your legal notice — and document it in a simple reference document. Then check every occurrence against it: Google Business Profile, your own website including legal notice and contact page, Bing Places, Apple Business Connect, directories, review portals, social profiles and association listings. Prioritise the sources with the greatest reach; the last five per cent of long-tail directories barely add anything.
5. Reviews: text and recency beat the star rating
Reviews were always important, but the logic has shifted. For a ranking system an average score is a conveniently usable number. For a language model the running text of the review is the genuinely valuable information: it says what work was done, how quickly there was a response, whether appointments held. Exactly those formulations turn up again later in AI answers.
One practical consequence follows: ask actively for reviews after a job — and in a way that lets customers name the specific service. A review containing "bathroom renovation in a period building, appointment kept, clean work" is considerably more valuable for your visibility on the query "who renovates bathrooms in period buildings?" than five stars without text.
Just as important: reply. Replies deliver additional text to the profile that you control, show activity, and let you name services and location cleanly once more. And because recency counts for more since the March update, a steady inflow of new reviews is worth more than a one-off collection drive. What you should not do goes without saying: bought or invented reviews are legally attackable and are detected increasingly reliably by the platforms.
6. Citations and portals: where AI systems get their data
A citation is any mention of your business with contact details on a third-party page — with or without a link. For AI systems citations are confirmation signals: the more independent sources give the same details, the more certain the assignment. Conversely, a single but prominent contradiction can tip a whole recommendation.
What matters most are the large directories and portals with their own data licensing, sector-specific platforms (chambers of trade, medical directories, restaurant portals), and regional media and association pages. How strongly individual portals come through in AI answers also depends on the providers' data deals — we described that in detail using OpenAI's Yelp cooperation as the example.
7. Your own website: structured data, location pages and llms.txt
Your own website is the only data source you fully own — and the one Ask Maps explicitly evaluates as well. Three components count here.
Structured data of type LocalBusiness
Clean JSON-LD markup with `LocalBusiness` (or the fitting subtype), `address`, `geo`, `openingHoursSpecification`, `telephone` and `areaServed` makes your core data machine-readable and unambiguous. It is the cheapest and most effective technical step there is — and it must contain exactly the same details as your profile.
Citable service and location pages
AI systems prefer to cite passages that answer a question directly. A service page that names the catchment area, scope of work, typical price range, process and response time in clear sentences is considerably more citable than a page full of marketing phrases. With several locations, each location needs its own page with its own address — not a combined page.
llms.txt: useful, but not a first-order lever
The idea behind llms.txt is appealing: a machine-readable markdown overview in the root directory that offers AI systems the most important pages and facts in compact form. The sober part belongs with it, though: no major AI provider has so far confirmed that it reads the file while crawling. We placed that in context in a separate article.
Our recommendation for local businesses: llms.txt is set up in half an hour and does no harm — so take it along, with a NAP block, a service overview and location pages. But it does not replace a maintained profile. If your time is limited, invest it in this order: profile, consistency, reviews, structured data — and then llms.txt.
8. Conclusion: data maintenance is the new local visibility
At first glance the numbers look discouraging: 1.2 % against 35.9 % sounds like a closed door. In fact the message is a different one. AI systems are not arbitrarily selective — they are strict. They preferentially recommend businesses whose data is complete, consistent and current, because only those businesses allow a recommendation with dependable contact details.
That is exactly where the opportunity lies for small and mid-sized companies. A well-maintained profile, consistent contact details and a steady flow of reviews are not a question of budget but of discipline. Anyone who does this homework while the competition is still discussing "GEO strategies" will appear disproportionately often on the short AI recommendation list.
Also worth reading: GEO: Generative Engine Optimization explained and Is your website visible in ChatGPT?
9. FAQ: the key questions about local AI visibility
Why does ChatGPT recommend so few local businesses?
AI systems are considerably more selective with local recommendations than classic search. The SOCi Local Visibility Index 2026 (over 350,000 locations, 2,751 brands) arrives at 1.2 % of locations recommended in ChatGPT against 35.9 % visibility in the Google 3-pack. AI models name only three to five businesses per answer and draw on the same data sources as local search — but with considerably higher demands on the completeness, consistency and recency of the data.
How important is NAP consistency for AI visibility?
Very. NAP stands for name, address, phone. If those details differ between the Google Business Profile, your own website and directory portals, a language model cannot merge the business into one clear entity. According to SOCi, contact details in AI answers are correct only around 68 % of the time — roughly every third local AI recommendation contains a wrong address, phone number or opening time. Consistency is therefore not a detail but the basic precondition.
Does an llms.txt file help local businesses?
Only to a limited extent. llms.txt is a proposal for a machine-readable overview file in a website's root directory. No major AI provider has so far confirmed that it reads the file while crawling. For local businesses it is therefore a cheap additional signal, but no substitute for a maintained Google Business Profile, clean NAP data and structured data of type LocalBusiness. If your time is limited, invest it first in profile, consistency and reviews.
What changed in 2026 at the Google Business Profile?
Three developments shape 2026. First, the Gemini-powered Ask Maps feature answers user questions directly from profile details, website content and review text. Second, since the core update in March 2026 Google weights the recency of reviews more heavily than their sheer number. Third, since early July 2026 Google has offered eligible profiles video verification as an alternative to the postcard. The common denominator: a maintained, active profile counts for more than one that was set up once.
