An AI assistant answering a local question does not rank websites. It assembles one recommendation from six data sources, and it favours the businesses those sources agree about. That makes local AI visibility mostly a data consistency problem, not a content problem, which is the opposite of how the rest of AI search works.
If you are still asking whether any of this helps a local business at all, the yes-or-no version is does AI SEO help local businesses. This page is the mechanics.
The six sources behind a local answer
Ask an assistant for a plumber in Sheffield, an accountant in Leeds, or a physiotherapist near a postcode, and something specific happens. The model is not retrieving a ranked list. It is reconciling several sources into a single confident statement, and each source carries different weight.
1. Map and place data. Google Business Profile above all, and Apple Business Connect and Bing Places behind it. This is the structural record of a trading location: name, address, category, hours, service area. Google AI Overviews and Gemini consume it directly. It is the foundation layer and it is free.
2. Directory and aggregator listings. Yell, Checkatrade, trade association registers, sector directories, chamber listings. Their individual authority is unremarkable. Their collective agreement is the point. Six listings saying the same address is corroboration. Six listings saying four different addresses is a reason for a model to name somebody else.
3. Review platforms. Google reviews, Trustpilot, sector platforms. Assistants read the text, not only the average, which changes what a good review is worth. More on this below.
4. Your own website. Contact details, service area, opening hours, LocalBusiness schema, and whether your pages answer the local question or just assert local presence.
5. Local media and community sources. Local press, council and BID pages, sponsorship listings, event pages. Sparse, but heavily weighted when present, because they are independent and geographically anchored.
6. Social profile data. Not the posting. The profile fields. An old address on a Facebook page is a contradicting source and it does measurable damage for no benefit.
Why consistency beats content locally
This is the part that surprises people who have done ordinary content marketing.
A model producing a local recommendation is making a factual claim about the physical world, and it will not do that confidently when its sources disagree. If your address appears three ways across the web, the model has a reconciliation problem. The cheapest resolution available to it is to recommend a business whose data is clean, or to give a category answer and list a directory.
So the highest-return local work is subtraction. You are removing contradictions, not adding material.
Run this audit. It takes an afternoon.
- Search your business name and note every result on the first three pages
- For each, record the name exactly as written, the address, the phone number and the category
- Flag every variation. “Ltd” versus “Limited” counts. A suite number present in one place and missing in another counts
- Fix the ones you control, in order of visibility
- For the ones you do not control, claim them or request the correction
Old addresses from a move, a previous trading name, and a phone number from before a merger are the three most common findings, and all three are worth more to fix than a month of publishing.
Reviews: what changed
Traditional local SEO treated reviews as a scored input. Star average, review count, recency.
Assistants read them. That is a different game, and it has a practical consequence: a review that says “great service, highly recommend” contributes almost nothing, because it contains no matchable information. A review that says “they rewired a 1930s semi in Crookes, worked around our kids, finished in three days and sorted the building control certificate” contains a service, a location, a property type, a constraint and a compliance outcome. That is five things a model can match to a query.
You cannot script reviews and you should not try. You can change what you ask for. “If you have a minute, it helps other people if you mention what the job was and roughly where” is legitimate, produces better reviews for humans too, and transforms what an assistant can do with them.
The same logic applies to your responses. A reply that repeats the service and the area is readable text on a high-authority third-party surface. Reply to everything, including the bad ones, factually and without defensiveness.
Proximity is not the only thing an assistant weighs
There is a persistent assumption that local means nearest. Assistants do respect proximity, but only in proportion to how much the query implies it matters.
For an emergency plumber, proximity dominates and there is little you can do to outrank geography.
For anything a customer will travel for, or receive remotely, the qualifier in the query does more work than the distance. “Accountant for landlords with a portfolio in Manchester” is answered by specialism first and location second. A specialist in Stockport will be named over a general practice in the city centre, because the model is matching the qualifier.
This is the single largest opportunity for local businesses that have any specialism at all, and it is under-exploited because most local marketing is still written to claim a town rather than to claim a problem.
Practically: write your pages around the qualified question rather than the geography. “Emergency electrician Sheffield” is contested and generic. “Landlord electrical safety certificates for HMOs in Sheffield” is winnable, specific, and attached to a customer who knows what they need.
Location pages, and how many you actually need
The templated town page is the standard local play and it is a poor fit here.
Thirty pages differing only by place name are near-duplicates. They dilute your entity, they give a model 30 weak signals instead of one strong one, and they carry a quality risk that has already caught plenty of businesses in ordinary search.
Build a location page only when you can put four things on it that are genuinely local: a real reason you serve that area, work you have actually done there, a local specific such as housing stock, a regulation, a major employer or a transport constraint, and local proof in the form of a review or a named job.
Two or three real ones beat 30 templated ones. If you serve a genuine region rather than a set of towns, one honest region page is better than pretending to be local in places you are not, and assistants are reasonably good at detecting the pretence because the rest of your data does not support it.
The schema that matters locally
LocalBusiness schema on the page that represents your actual premises, with the address, opening
hours, service area and telephone number matching your Google Business Profile exactly. Not
approximately. Exactly.
One caution worth stating because it is a common and damaging error: do not put LocalBusiness
markup on pages for towns where you have no premises. That is a false claim about the physical
world, it is machine-checkable, and it undermines the credibility of the rest of your markup. Use
areaServed for coverage and reserve LocalBusiness for places you actually are. The full
treatment is in our
schema markup guide.
A first 60 days for a local business
| Week | Work | Why it is in this position |
|---|---|---|
| 1 | Full listing audit, every variation recorded | You cannot fix contradictions you have not found |
| 2 | Fix everything you control, claim what you do not | Highest return per hour of anything on this page |
| 2 | Google Business Profile finished properly: services, hours, area, 20-plus photos, self-answered questions | Free, direct input to Overviews and Gemini |
| 3 | LocalBusiness schema matching the profile exactly |
Machine-readable confirmation of what you just aligned |
| 3 | Confirm AI crawlers are not blocked in robots.txt | Cheap, and a surprising number of sites fail it |
| 4 to 6 | Review request process changed to ask for specifics | Compounds for years, costs nothing |
| 4 to 8 | Two or three real location or qualified-service pages | Specificity, not coverage |
| 8 | Test 20 local prompts across 5 assistants, record | Your baseline, and the only way to know later |
Nothing there requires a retainer. It requires somebody to actually do it, which is the part that usually fails.
Sectors where this works hardest
Local plus regulated is the strongest combination, because the compliance questions are specific, the answers are checkable and almost nobody publishes them properly. See GEO for dental practices, GEO for private healthcare clinics, GEO for accountancy practices and GEO for construction companies.
City-level detail sits on the location pages: Sheffield, Leeds, Manchester, Birmingham and the rest are indexed from GEO agency UK.
For proof that the shape works, a dental directory went from zero to 66,000 impressions on local-first data and question work. MarGen’s pricing is published.
AI Search for Local Business: Common Questions
How do AI assistants decide which local business to recommend?
They assemble an answer from map and place data, directory listings, review platforms, your own site and any local press, then weight sources that agree with each other. Consistency across those sources matters more than the quality of any single one, which is why a business with an outdated address in three directories underperforms a weaker competitor with clean data.
Does an AI assistant use Google Business Profile data?
Google AI Overviews and Gemini use it directly, and it is a strong signal for the rest by proxy, because it feeds the wider web that other models crawl. A complete profile with services, hours, service area, photographs and self-answered questions is the single highest-return asset a local business owns, and it is free.
Do I need to be near the customer to be recommended?
For genuinely proximity-driven trades, yes, and assistants respect that. For services people will travel for or receive remotely, proximity matters much less than specificity. A specialist 40 miles away gets recommended over a generalist round the corner when the question contains a qualifier the specialist answers and the generalist does not.
How important are reviews for AI search visibility?
More important than for traditional local ranking, and for a different reason. Assistants read review text, not just the star average. Reviews that name the specific service, the location and the problem solved give a model language it can match to a query. Twenty detailed reviews beat 200 that say great service.
Why does my competitor get named in ChatGPT and I do not?
Usually one of three things. Their entity resolves cleanly and yours contradicts itself across sources. They are listed where the model is looking and you are not. Or their site answers the specific question and yours has a services page that answers it vaguely. It is rarely about who is better at the work.
Do I need separate pages for every town I serve?
Only where you can say something genuinely different about each one. A town page with the name swapped is a thin near-duplicate and it dilutes rather than helps. Two or three real location pages with local specifics, local proof and a genuine reason to exist outperform 30 templated ones, and they will not attract a quality problem later.
How long does local AI visibility take?
Faster than most AI search work, because the levers are data corrections rather than authority building. Cleaning up inconsistent listings and finishing a Google Business Profile can show up within four to eight weeks. Review depth and local press build over three to six months.
Does social media affect local AI recommendations?
Indirectly and modestly. Its real job is consistency, not reach. A Facebook page with the old address on it is a contradicting source, and removing that contradiction is worth more than any amount of posting. Get the profile data right, then treat posting as a separate question about customers rather than about models.
Where to Go Next
The off-site layer this page depends on: ecosystem validation for AI search · review platforms and trust signals · social platform consistency
Adjacent situations: AI search for small business · AI search for retail · does AI SEO help local businesses
Diagnose the specific problem: why am I not in AI search results · competitors in AI answers and not you · AI giving wrong information about my company
Have it delivered: MarGen is a UK GEO agency with published pricing, a published method and documented results.