What ChatGPT Says When Someone Asks for an MSP in Your City

We asked ChatGPT the question a business owner actually asks before hiring an IT provider — "I run a small business near [city], who should manage our IT?" — across eight US markets where we knew a capable managed services provider operates. Six of those eight providers were not mentioned at all. Other firms were, and the reasons they got in were specific, visible and mostly fixable.

This is what we found, and how to run the same test on your own market in about twenty minutes.

The short version

• Six of eight established MSPs were absent from the answer to their own buying question.

• Metro-level answers ran on review volume and directory listings — not on how good the firm is, and not on how well it ranks.

• Niche answers ran on self-description. One firm serves a specialty and was invisible in it, purely because competitors said the words and it didn't.

• A single listing can carry an answer — a chamber of commerce entry, a compliance directory, a professional association endorsement.

• Small markets are winnable outright. One firm was named first with no reviews at all.

The test, exactly as we ran it

The prompt matters more than anything else, in two ways.

Use the buyer's words, not the industry's. "Best managed service provider" is how vendors talk. "I run a small business near [city] — who should manage our IT?" is how buyers talk, and the two produce noticeably different answers. We used the second.

Match the real geography. This is the one that bit us. We ran a prompt using a regional description for one firm's territory, and the model interpreted the region as a different corridor a hundred miles away — returning a competitor set from a market the firm doesn't serve. We nearly acted on it. If you test at regional level, check that the answer names places you actually cover before you believe a word of it.

We ran each market's prompt fresh, recorded which firms were named, and recorded which sources were cited underneath.

What decided who got named

In metro markets: review volume and directories.

The firms named in metro-level answers were consistently the ones with the most Google reviews and the most aggregator listings. In one market the named competitors carried 156 and 138 reviews; in another, 85 and 92. The answers leaned on aggregator pages — Expertise, Clutch, and Google Business Profiles, which ChatGPT now renders inline as a map.

Note what is not in that list: page-one rankings, site quality, years in business, or how good the work is. The model was answering a recommendation question by consulting the places that publish recommendations, which is a reasonable thing to do and has nothing to do with SEO as most firms practise it.

In niche markets: how you describe yourself.

The clearest single finding. One of the eight specialises in IT for dental practices — genuinely, it is their business. They did not appear in the answer to "who handles IT for dental practices near [city]." The three firms that did appear all had one thing in common: they say "dental IT" explicitly and repeatedly on their own websites.

That is not an authority problem. It is not fixable with links or reviews. It is a positioning problem, and the fix is writing down the thing you already are, in the words a buyer would use, in a place a machine can read.

Sometimes, one citation is enough.

In three of the eight markets, a competitor's inclusion traced to a single source: a compliance directory listing showing a certification level, a county medical association endorsement, a local chamber of commerce entry. Not a link-building campaign — one listing, in a place that happened to be the authority the model reached for.

And one firm got quoted because it published its prices.

In one market a competitor was not only named but quoted, with its per-user pricing repeated in the answer. Every other firm in that market keeps pricing behind a contact form. The model could not quote what it could not read, so it quoted the one firm it could.

That is worth sitting with if your pricing lives behind a form. You may have good reasons for that. The cost is that you are unquotable.

The two that did appear

One was named first in its home market, for both of its office towns, on the strength of its own website plus a chamber of commerce listing. It has no reviews to speak of. In a market thin enough that no competitor has built a directory presence, being clearly described and locally listed was enough to win outright.

That position is real, and it is fragile. It rests on two sources. Any competitor who collects twenty reviews takes it.

One was split. Named on the hyper-local prompt — "small business near [suburb]" — and absent from the regional one covering the wider metro. The regional answer was the review-volume answer. The local one wasn't.

If you only test at one level of geography, you will get a misleading picture of where you stand.

What this means if you run an IT firm

Three things follow, and none of them are what the AI search industry usually sells.

First, your rankings are not the problem and not the answer. Several of these firms rank respectably. It didn't help. Independent research points the same way: Ahrefs, analysing 15,000 prompts, found that only 12% of URLs cited by ChatGPT, Gemini and Copilot appear in Google's top 10 for the same prompt. You can win the search results and lose the answer.

Second, the levers here are unglamorous. Reviews. Directory and association listings. Saying plainly what you do and where. Publishing information rather than gating it. There is no technical trick in any of the eight results we looked at — they are the same foundations AI SEO rests on generally.

Third, the gap is checkable. You do not need a tool or a retainer to find out whether this is happening to you. You need twenty minutes.

How to run this on your own market

1. Write five to fifteen prompts in your buyer's voice. Not "best MSP" — "our server keeps going down, who do we call in [city]", "who manages IT for law firms near [city]", "I run a small business near [suburb], who should handle our IT?" Include at least one hyper-local and one regional.

2. Run them in ChatGPT, Perplexity and Google. Same day. Fresh chats.

3. Record three columns: were you named, was your site cited, and who else was named. The third column is the one people skip and the one that tells you what the answer is actually running on.

4. Check the sources underneath. This is where the real finding is. If the answer leans on a directory, that directory is your route in. If it leans on reviews, you have a review problem, not a content problem.

5. Sanity-check the answer before you act on it. Confirm the named firms exist and really serve that market. We caught one result that inserted a business with no business being there. These systems are confident and occasionally wrong, and a plan built on a hallucinated competitor set is worse than no plan.

6. Write down the date and keep the list. Re-run the same prompts monthly. The value is entirely in the comparison — a single snapshot tells you where you stand, a series tells you whether anything you did worked.

The uncomfortable part

If you run this test and find you're absent, there is no dashboard that was going to tell you. No impression, no session, no referrer — the prospect asked, got three names that weren't yours, and the entire event is invisible to your analytics.

That is the real argument for checking manually. Not because manual is better, but because nothing else is currently looking.

Want us to run it for your market?

That is the first section of our written SEO and AI visibility audit. We do this work for IT resellers and managed service providers: your prompts, your market, who gets named, what the answers are running on, and what we'd fix first. It's free, and you get a document rather than a sales call.

Get a free audit →