We asked ChatGPT for the best AI SEO agency for MSPs five times. Our own page got us named twice, and last both times

One prompt, typed exactly as a buyer might: best ai seo agency for MSPs.

• ChatGPT temporary chat, signed in, with Unpersonalized mode on. That mode tells ChatGPT to ignore memory, plugins and custom instructions.
• A fresh chat for every run. No follow-up questions.
• ChatGPT used web search on every run without us asking.
• For each run we logged whether GeoCited was named, its position, who was named instead, and every page ChatGPT cited.

We ran it six times and counted five. The reason is below, and it matters more than it sounds.

On the first run, the page said the chat would ignore "plugins and custom instructions". It did not say memory. The next time we opened a temporary chat, the setting had switched to Personalized. We changed it to Unpersonalized and kept it there for every run we counted.

That first run put GeoCited first of seven. It then closed with an offer to research "competitors for GeoCited", which only makes sense if ChatGPT knew who was asking.

So the only run that ranked us first is the one we can't trust. If an agency shows you a screenshot of ChatGPT putting them at number one, ask what mode it was taken in. Ours would have looked great.

GeoCited was named in 2 of 5 runs and was last on the shortlist both times.

How often each agency made the shortlist:

Twelve agencies appeared across five runs. No agency made all five. Agency A came closest.

ChatGPT cited 12 distinct pages across the five runs.

A page on Agency E's site was cited in all five runs.

Look at runs 2 and 5. ChatGPT built both answers almost entirely from roundup articles: Agency E's own roundup, a second roundup (Roundup 2), Western News and a third-party study. It did not open a single agency's own MSP page on either run.

We opened the three roundups. None of them mentions GeoCited:

• Agency E's list is an MSP SEO agency's own roundup, updated 1 August 2026. It ranks twelve agencies and puts Agency E first. It says so openly: "[Agency E] built this list and appears on it." Agency A is second.
• Roundup 2, updated 4 August 2026, ranks nine agencies and leaves its own publisher off. Agency A is first.
• Western News, dated 3 September 2026, lists six agencies. Agency A is first.

That explains Agency A. It made the shortlist in 4 of 5 runs, but ChatGPT cited Agency A's own page only once. Its place in the answer comes from other people's pages. Ours can only come from our own page, because nobody else has written about us for this question yet.

When ChatGPT answers from roundups, we are not in the answer. When it reads agency pages, we sometimes are.

Getting read was enough to be named. It was not enough to be ranked.

Here is how ChatGPT described us when it did name us:

"GeoCited — unusually focused specifically on AI-search/GEO for MSPs rather than selling a broader marketing stack." (Run 1)

"GeoCited — Interesting if your only objective is 'make AI assistants mention my MSP when prospects ask who to hire.' Unlike full-service agencies, it says AI SEO/GEO is its sole service." (Run 3)

Both descriptions were accurate at the time. Our MSP page said we sold one thing and listed what we didn't do. ChatGPT picked that up.

The trouble is that ChatGPT's own buyer advice, repeated in almost every run, asks for the one thing our MSP page cannot show yet:

• "Actual MSP case studies" (Run 1)
• "Actual MSP case studies with attributable leads/revenue" (Run 3)
• "2–3 recent MSP case studies" (Run 4)
• "MSP case studies, AI SEO methodology, pricing, reviews, and evidence of actual results" (Run 5)

Our MSP page has no MSP client results. In run 4, ChatGPT described Agency E as offering MSP SEO and AI search "with published MSP case studies and measurable outcomes." Being specialised gets us described. MSP proof is what ChatGPT keeps asking for, and our page has none yet.

ChatGPT also wanted local SEO in the mix. It put local SEO on its checklist in 4 of 5 runs (1, 2, 3 and 5), and in run 3 it asked how an agency's AI work fits with "local SEO + Google Business Profile + service/location pages". Our MSP page said plainly that we didn't do local SEO, and that cost us with any buyer who read the answer literally. We have since rewritten that section to explain how the AI work fits with your SEO.

1. 01Ask how many runs. One run of this prompt showed us fifth, another sixth, and three left us out. All five happened on the same evening.
2. 02Ask what mode the screenshot was taken in. The only run that put us first was the one where we could not confirm memory was off.
3. 03Ask which pages the answer came from. An agency that shows up only through a list it wrote itself is a different case from one that independent lists include.
4. 04Ask for MSP results, not AI results. ChatGPT asks for them. You should too.

• Getting onto the roundups ChatGPT actually reads. Roundup 2 leaves its own publisher off the list and judges agencies on AI search, so it is our first request.
• Putting MSP proof on the MSP page as soon as an MSP client agrees to be named. No invented numbers until then.
• Running this prompt again every week under the same protocol, including the mode setting, and adding each new set of runs to this page with its date.

• Five runs, one evening, one account, one location, one engine. It is not a share-of-answer rate.
• It does not show that roundup placement will get us named. It shows that we are missing from the pages that decide most runs.
• It says nothing about Perplexity, Gemini, Claude or Google AI Overviews.
• Where we quote ChatGPT on other agencies, those are its claims, not ours.