Key takeaways
- Brand-level AI trackers average everything. A business can win in one pocket of a city and vanish in the next, and the average hides it.
- Start with a free manual test: ask ChatGPT, Perplexity and Gemini the same three customer questions, phrased for each neighborhood, and write down who is named and which sites are cited.
- Use a map geo-grid to see the Google Maps side. Green is a top-3 result, red is nowhere. Then compare it with what AI says, because the two often disagree.
- Treat the location, not the brand, as your reporting unit, and repeat the test across many prompts. One prompt tells you very little.
- Paid tools exist for both halves of the job, but most published rankings are vendor blogs. Check prices on the vendors' own pages.
Why you can be visible in one place and missing in another
Say you run a dental clinic, a plumbing company or a coffee shop group. You search for yourself and look fine. Your reviews are good. Then a customer two neighborhoods over asks an AI assistant for the best option nearby and you are not in the answer.
This is not a glitch. Local results are built from place-specific signals: where the searcher is, which listings sit close by, which pages mention a neighborhood, and which reviews describe that area. A business that is strong in one pocket can be weak a few miles away, and a single brand-wide score smooths that over.
One of the sources I looked at puts it well. A home-services company can win an emergency plumbing question in one market and lose a roofing comparison in another, so the location, not the brand average, should be the reporting unit (Cheers). I agree with that framing. If your report has one number for the whole company, it is probably hiding the problem.
A caveat before going on. I could not find official documentation from OpenAI or Google on how their assistants use a user's location, and I found no independent study of neighborhood-level variance in AI recommendations. So treat the mechanics below as a working model and verify with your own tests.
Step 1: Run the free manual test
Before you pay for anything, spend an hour on this. The idea comes from general-purpose checks like the ones Neil Patel shared: ask ChatGPT who you are, ask it what your customers ask, and ask several AIs, not one.
Those checks say nothing about location, so add that layer yourself.
The neighborhood version
- Pick your three most valuable customer questions. For example: "best emergency plumber", "who fixes a water heater same day", "affordable family dentist".
- Make one version of each question per area you serve. Use the neighborhood or suburb name, and also try "near me" with your device location set there if you can.
- Ask ChatGPT, Perplexity and Gemini. Use a fresh chat each time so earlier context does not leak in.
- Log the answer in a sheet: area, assistant, prompt, businesses named, your position, and which sites were cited.
- Repeat monthly. A speaklocal.com comparison suggests the same routine for a single location: three questions, three assistants, note who is named and what is cited.
A simple log looks like this:
| Area | Assistant | Prompt | You named? | Position | Top cited sources |
|---|---|---|---|---|---|
| Northside | ChatGPT | best family dentist in Northside | Yes | 2nd | Yelp, clinic site |
| Northside | Gemini | best family dentist in Northside | Yes | 1st | Google Maps listing |
| Riverside | ChatGPT | best family dentist in Riverside | No | n/a | Healthgrades, competitor sites |
| Riverside | Gemini | best family dentist in Riverside | No | n/a | Google Maps listing |
The interesting rows are the ones where you are present in one area and absent in another. The cited-sources column tells you why. If a competitor in Riverside is cited from a directory you are not listed on, you have found your first fix.
Don't trust one run
AI answers vary. One prompt tells you little. Promptwatch Data looked at how many domains ChatGPT cites per response (Sep 28 to Oct 4, 2026) and found that 47.6% of responses with citations cite 10 or more unique domains, and organic prompts with no brand named are the most crowded, at 52.5%. Brand-specific prompts draw on fewer sources. That report is about general ChatGPT behavior, not local, but the lesson transfers: with that many sources in play, you need many prompts per area before you believe a pattern.
If you can, run 10 to 20 phrasings per neighborhood. Mix organic prompts (no brand named) with comparison prompts. The organic ones are where you find out whether you get recommended at all.
Step 2: Check the map side with a geo-grid
AI answers are one half. Google Maps is the other, and the two often recommend different businesses. A geo-grid shows your map ranking at many points around a location instead of one.
How to read one: green means you show in the top 3 on Google Maps at that point, and red means you do not show at all. A business can rank well in one small pocket and nowhere else, which is the exact problem in this guide. (That description comes from a Facebook group post, so it is not authoritative, but it matches how every geo-grid tool displays results.)
A few practical notes:
- Grid sizes run from 3x3 up to 21x21 in some tools. Bigger grids cost more.
- Credit-based tools charge per map point. One credit is one point, so a 5x5 grid uses 25 credits and a 7x7 grid uses 49. According to AuraLocal's comparison, keeping 10 keywords refreshed daily on a 7x7 grid uses about 14,700 credits a month per location. Daily large grids get expensive fast.
- Weekly scans on a modest grid are enough for most small businesses.
Local Dominator published a study (vendor data, not independent) saying mobile and desktop returned the same businesses in the same order 99.85% of the time across 40 keywords, 25 US metros and 703 locations. If true, you do not need to scan both devices separately. I would not bet a budget on one vendor's number, though.
Step 3: Compare the two pictures
Put the geo-grid next to your AI log. You will usually land in one of four situations:
| Maps | AI answers | What it usually means | First thing to try |
|---|---|---|---|
| Green | Named | Healthy area | Protect it, keep reviews fresh |
| Green | Not named | Strong listing, weak web footprint | Add pages and third-party mentions for that area |
| Red | Named | AI is pulling from other sources like directories or articles | Fix your listing and categories |
| Red | Not named | Truly invisible there | Build local content, reviews and citations for that area |
The second row is the one people miss. Your Google Business Profile can be fine while ChatGPT never mentions you because nothing on the open web ties your brand to that neighborhood.
Step 4: Figure out why one area is weak
Neighborhood gaps usually trace back to a short list of causes.
- No page or mention for the area. If your site only says "serving the metro area", an assistant has little to match against "in Riverside". Dedicated, genuinely useful location pages help. Thin copy-paste pages do not.
- Reviews that never mention the area or service. Local search guidance from Straight North points out that people check reviews before visiting and that reviews affect local ratings. Reviews that name the service and place give both Maps and AI more to work with.
- Listings that disagree. Different names, phone numbers or addresses across directories confuse everything. Tools like Yext are built to keep listings consistent.
- Competitors cited from sources you are absent from. Look at the cited-sources column in your log. If the same two directories or local news sites keep showing up in the weak areas, get listed or mentioned there.
- Crawler access. If AI crawlers cannot read your location pages, you will not be retrieved. Check your robots rules and server logs.
One inference of mine, not from any report: since AI assistants run web searches behind the scenes (Promptwatch's query fanout data shows the average number of searches per response moving around 1 to 2 in recent months), location words in those queries, like a neighborhood name or "near me", probably decide which local pages get retrieved. Write headings that read like the query and put the entity and category first, as that report suggests.
Tools that help
Tools fall into two groups, and you probably need one from each if you operate in several areas.
Geo-grid and local AI visibility tools
These measure by location. Prices are as reported in October 2026 and vendors change them often, so confirm on their sites. Most of the comparison pages I read are written by vendors who rank themselves well, so read them with that in mind.
| Tool | Reported starting price | What it does | Watch out for |
|---|---|---|---|
| Local Falcon | Credit packs from $24.99/mo | Grids from 3x3 to 21x21, Google and Apple Maps, AI scans across ChatGPT, Gemini, AI Overviews, AI Mode, Grok | Monthly credits expire; daily big grids burn credits |
| BrightLocal | Track plan from $41/mo per location | Local AI visibility view inside each location, tracks ChatGPT and Google AI | Priced per location |
| Speak Local | From $49/mo | Asks ChatGPT, Perplexity and Gemini directly, grid up to 7x7 | About 5 questions tracked weekly on the base plan |
| AuraLocal | $39 for 1 profile | Flat per-profile pricing, 10 keywords, 24-hour refresh | One 7x7 grid size, no Apple Maps |
| Local Dominator | $39/mo for 5,000 credits | Geo-grid, profile tools, AI tracker | Credit-based |
A note on accuracy. Many AI grid scans sample the model rather than reproduce what a real person standing on that corner would see, so treat the results as estimates. I could not find a primary source on exactly how ChatGPT or Gemini use location, so I would not claim more precision than that.
Brand-level AI visibility platforms
These are not location-specific by default, but they are good at what the manual test cannot do: tracking many prompts daily, comparing competitors, and showing which pages get cited. Pair them with a geo-grid tool.
Promptwatch is the one I would look at first if you also want to act on what you find. Its monitors can be pinned to exact coordinates (a 20 m to 5 km accuracy radius), so ChatGPT, Google AI Overviews and AI Mode answer as if asked from that spot. That is built for "near me" queries and multi-location brands. It sits on the Business plan and above, so check promptwatch.com/pricing. It also has crawler logs, a content gap view and a Content Agent for publishing, which matter once you know which neighborhood is weak.
Other options in the same family include Profound, which has enterprise pricing and publishes region limits by plan, and cheaper trackers like Otterly and Peec.
Semrush has a local product and an AI visibility toolkit, reported at $99 per domain per month for the latter.
For listing consistency across directories, Yext is in our catalog.
For reviews and reputation by location, Birdeye is another option.
If you are comparing options, the GEO software directory at bestgeosoftware.com lists many more.
Here is how I would choose:
| Situation | What I would use |
|---|---|
| One location, tight budget | Manual test monthly plus a low-cost geo-grid tool |
| A handful of locations | A local tool such as BrightLocal or Speak Local, plus a manual spot check |
| Many locations or franchises | A platform with location-pinned monitoring, plus a geo-grid tool for Maps |
| Agency with local clients | Per-location reporting, white-label output and a client view |
Step 5: Turn the findings into a fix list
Once you can see the gap, make a list per weak area. Keep it small.
- Pick the two or three worst neighborhoods by revenue potential, not by size of the red patch.
- For each, list the sources AI cited for competitors and note which ones you are missing from.
- Fix listings and categories first. It is cheap and quick.
- Publish one real page per area: services, local proof, staff, photos, answers to the questions people actually ask there. Straight North's advice is to find what locals ask and prepare real answers from your work. If it is a repeated question, it is on people's minds.
- Ask happy customers in that area for reviews that name the service.
- Re-run the same prompts in 4 to 6 weeks. Same wording, same assistants, same log.
Do not expect AI answers to move overnight. Promptwatch's content half-life data (pages ChatGPT cited on July 6, 2026, tracked for 90 days) found landing pages and docs stayed cited longest, while news lost half its pages within 36 days. Durable landing pages for each area are a better bet than a one-off post.
Common mistakes
- Reporting one brand-wide visibility score and calling it done.
- Testing from your own office and assuming the result holds across the city.
- Running a prompt once and drawing conclusions.
- Paying for daily large grids when weekly small ones would show the same pattern.
- Taking vendor comparison rankings at face value. Nearly all of the ones I found come from the vendors themselves.
- Ignoring the Maps half because AI feels newer. Your customers use both.
Where to start this week
Pick your three best customer questions and your four most important neighborhoods. Run the twelve prompts across ChatGPT, Perplexity and Gemini, log the results, and run one geo-grid scan for your main service. If one area is clearly red or absent, you have your answer and your first project. Then decide whether the pattern is big enough, across enough locations, to justify a paid tool. If you operate in five areas, probably not yet. If you operate in fifty, definitely.
For more on how AI sources get picked in general, the research hub at promptwatch.com/data has reports you can read for free.