Tonight, open ChatGPT, Perplexity and Google's artificial intelligence (AI) answers, and ask each one the question your next client is asking: who is a good managed information technology (IT) provider in your city. Ask three times in each, in a fresh chat each time. Nine answers. Count how many name your firm, and write down the names that show up instead.
If you own a managed service provider (MSP) and your count is low, open the websites of the firms that did come up. Ask what they have that you do not. The answer may not be on their homepage at all.
An answer engine, meaning ChatGPT, Perplexity or the AI answers at the top of Google, does not pick the firm with the most polished homepage. By my reading, it picks the firms that other pages keep naming. The fastest way I know to become one of those names is to publish lists about other people. Your own site still has to carry the basic facts about you, and I will come back to that, because it is the half most advice gets right.
For ten years my job was pricing businesses from the buyer's side, and today I am fractional Chief Revenue Officer (CRO) at a national MSP in the United States, so where new clients come from is my number.
If you want the own-site half first, read should an MSP put pricing on its website. If you want to see a list about other people that I publish, look at the best MSP podcasts.
How an answer gets made
When a buyer types something like "best IT support company near me" into one of these tools, the engine usually does not answer from memory alone. It does retrieval: it runs searches, reads the pages that come back, and writes an answer from what it read. Google's own documentation for its AI features says AI Overviews and AI Mode may issue several related searches across subtopics and sources to build a single response. One question, many searches, many pages.
As I read it, the engine starts from roughly what sits on page one of search results. Then it looks for more evidence: is this firm mentioned on Reddit, on LinkedIn, on YouTube, in other publications, in links from sites it does not own? It discounts what looks fake or spammy. Then it settles on a handful of names. That is my model, simplified, and none of the engines publishes its full recipe. But it matches what I see in the results. Smaller firms that publish fresh material and show up elsewhere often beat legacy firms that have gone quiet.
The word for that "is it mentioned elsewhere" check is corroboration: the same name showing up on pages the firm does not own. A firm that describes itself well on its own site has made one claim. A firm that is named on a local business roundup, a podcast page, a LinkedIn post and a list article has been confirmed by sources it does not control. Agreement across independent sources is what these engines appear to look for before they name a brand with confidence.
The firm describes itself
one claim, from a source it controls
Other pages name the firm
confirmed by sources it does not control
Before you trust your nine answers, know that the lists change from run to run. A study by SparkToro in January 2026, run with 600 volunteers across ChatGPT, Claude and Google's AI Overviews, found that if you ask ChatGPT or Google's AI the same thing 100 times, there is less than a 1 in 100 chance that any two responses give the same list of brands, and roughly 1 in 1,000 that two lists come back in the same order. That study covered consumer products, not IT services, but the lesson carries: there is no "number one in ChatGPT". What you can measure is share of answers, meaning how often your name appears across many repeated runs of the same prompts. That is why the test above asks three times, not once.
Why your site is not the answer
Most of the MSP sites I have audited read the same. A services page listing managed IT, cybersecurity, cloud and backup. A "why choose us" section about responsiveness. A blog with posts on phishing and password hygiene. Nothing on the page tells an engine, or a buyer, why this firm and not the next one down the results. Content is easier to produce than it has ever been, and harder to get results from, because everyone else can produce it too.
The blog is where this gets expensive without anyone noticing. If your blog is filled by a marketing program that sends the same posts, lightly reworded, to many member firms, you have syndicated content: the same post published on many sites.
You can check this in two minutes. Take the exact title of your most recent blog post. Search it on Google inside quotation marks, and add a space and -site:yourdomain.com (with your own domain) so your site drops out. Count how many other MSP sites carry the same headline. If the answer is more than a couple, my view is that the post adds nothing an engine can use to tell your firm apart, no matter how well written it is.
None of that makes the program a bad buy. Those programs are built for mail, email and referrals, for staying in front of the clients and referral partners you already have, and they can do that job well. They are not built to make your firm the corroborated name in your city. That is a separate layer, and it is the one the kit misses.
Your site still has a job
It would be convenient to stop here and say your website does not matter. It does, and the evidence says so.
A Yext study of 6.8 million citations in answers from Gemini, OpenAI and Perplexity found that 86% of the sources the engines linked to were managed by the brands themselves: 44% websites and 42% listings, such as business profiles and directory entries. Reddit was about 2%. Two limits on that. The study covered retail, finance, healthcare and food service, not IT services. And the data is Yext's own, from a company that sells listings management, so it has an interest in the answer.
Even so, I take the point. Your own site and your listings are where the engines go to get facts about you, such as what you do and where you are. Whether you are among the few firms named for "best managed IT provider in [your city]" is a different question. That depends on how many independent pages put you on a list. Citations and recommendations are not the same thing, and you need both.
I have audited MSPs whose own site barely figures in what the engines say about them. If your site does not state the basics clearly, the engine has to take them from somewhere else. So the first job is dull and worth doing: make your name, address, service list and the names of your leadership team match everywhere they appear. Your website, your Google business profile, LinkedIn, your chamber of commerce listing, any directory you are on. If an engine finds three versions of your address, it has to guess.
The list about other people
Here is the move I would make. Publish list articles, meaning pages built around a ranked or grouped list of named options, about people and things your buyer cares about, where you are not the subject.
The mechanism is corroboration you create. A good list about other people can rank on page one for questions buyers actually search, which puts your site in the set of pages the engines read when they build an answer. The people you name tend to notice. If they share it, link to it or mention you back, each of those is another page, one you do not own, that names you. Your homepage cannot do that for itself. A list can, because it is useful to someone other than you.
An MSP I work with went from about 260 to about 10,000 estimated organic visitors a month in nine months. Organic means from unpaid search results. Most of that traffic landed on list articles about other people's tools and firms. That is search traffic, not a measure of AI answers, and I would not claim an AI number from it. But page one is where the engines start, on the model above, and those lists are what put that firm there.
about 260
estimated organic visitors a month, at the start
about 10,000
estimated organic visitors a month, nine months later, most of it on list articles about other people's tools and firms
search traffic, not a measure of AI answers
I do the same thing on this site. The podcast list linked above checks every MSP podcast feed and reports which are still publishing, slowing or gone dark. A second one covers books for MSP owners. Neither is about me, which is the point.
The rules I hold these lists to:
- Lean toward named people. On one kind of list my agency publishes, the rule is named individuals only, no firms, because a person notices being listed in a way a logo does not, and the list becomes the start of a relationship.
- Never list a firm with "MSP" in its name. That rule holds on every list my agency publishes.
- Every sentence has to matter to the business owner reading it. If a paragraph exists to show you did research, cut it.
- Verify every fact about every named person or company before you publish it. If you cannot prove something, say you could not, or leave it out.
For an MSP that sells to small businesses, the lists write themselves once you think like your buyer. The practice management tools a dental office in your area can choose from. The cyber insurance brokers who understand small firms. The local business groups worth joining. Those are hypothetical topics, and which ones fit depends on your verticals, but the shape holds: useful to your buyer, and about somebody else.
How to score it
The nine answers you collected tonight are a starting point, not a score. A single screenshot proves nothing, because the next run will give a different list.
The version I use is simple enough to run yourself. Write a fixed set of prompts and keep it fixed. Split them into branded prompts, which mention your firm by name (for example, "is [your firm] a good IT provider?"), and non-branded ones, which do not ("best managed IT for a law firm in [your city]"). The non-branded set is what a buyer who has never heard of you types. Run each prompt several times on each engine. Count your share of answers against one named competitor, the firm you lose deals to. Then look at which pages the engine cited for each prompt, because that tells you which lists you are missing from and which ones you could write.
Run the same set again in a month and compare. The change between the two is your number.
If you hire someone to do this, ask for the prompt set, how many times each prompt was run, which competitor they benchmark you against, and the pages cited per prompt. A vendor who shows you one screenshot, or promises you will "rank first in ChatGPT", is selling something the engines do not offer. Nobody can promise that an engine will name you. What you can do is make it more likely and measure whether the share is moving.
What to do this month
Run the test properly: ten prompts, a mix of branded and non-branded, three runs each across ChatGPT, Perplexity and Google's AI answers, with one named competitor as your benchmark. Write down the share for both of you.
Search the exact title of your last blog post in quotes, minus your own domain, and count the copies. That tells you whether your blog does anything to set you apart.
Fix the facts. Name, address, services and leadership, the same everywhere they appear.
Then publish one list about other people your buyer cares about, held to the four rules above. Run the prompts again next month and see whether the share moved.
If you want to work through your own result with other owners asking the same question, Know Your Number is a private room for MSP owners with a regular call that covers website, search and AI visibility.