Blog AI search and provider discovery
AI search and provider discovery
What AI Search Tells B2B Buyers When They Look for a Provider
We asked four AI search products 20 questions a B2B buyer might ask on the way to hiring a managed IT or cybersecurity provider. Almost every answer named a type of service. None of the 80 answers explicitly recommended a specific firm.

In a September 2026 capture of 80 answers (20 buyer questions in each of four AI search products), 79 named a service category and 56 named managed IT or an MSP, but 0 explicitly recommended a specific firm, including all 24 provider-choice answers. 42 of 80 answers showed source links, and 28% of the 424 visible links went to government sites.
79 of 80
answers named a type of service, and 56 of those 80 named managed IT or an MSP
0 of 80
answers explicitly recommended a specific firm, including all 24 answers to questions about choosing a provider
42 of 80
answers showed at least one source link, from 20 of 20 in Perplexity to 0 of 20 in Gemini
The thesis
In this capture, AI search did the category work for the buyer, not the vendor pick. It told people what kind of help they needed and how to judge it. The choice of firm was left to them.
The scene
A buyer asks how to shortlist providers
One question from the set, and what came back.
Picture an operations lead at a Toronto manufacturer. Patching is behind, backups are untested, and the one IT generalist is stretched thin. Before calling anyone, she types a question into an AI search product: "How should a Toronto manufacturer evaluate and shortlist managed IT providers before signing a contract?"
She asked how to shortlist providers, not who to call, and that is what came back: what to look for, what to ask, which warning signs to walk away from. The answer names the kind of help she needs. It does not name a firm to call.
No answer in the capture named a firm as a pick, in any product or at any stage.
What we asked
Twenty buyer questions, four AI search products
The path a buyer takes from a problem to a provider.
We wrote 20 questions in plain buyer language, with no brand names in them, about managed IT and cybersecurity for small and mid-sized B2B companies. They follow the order a buyer usually thinks in:
- Understanding the problem (7 questions), for example why a small team keeps losing hours to unresolved IT tickets.
- Choosing a type of service (7 questions), for example when a company needs a managed service provider versus staying in-house.
- Choosing a provider (6 questions), for example what to ask before hiring an MSP that will touch clinical systems.
Some questions were worded for Canadian buyers, some for US buyers and some for neither. We put all 20 to ChatGPT, Perplexity, Gemini and Google AI Overviews in September 2026. That gave 80 answers, and we read every one of them.
For each answer we tracked four separate things, because they are easy to blur: whether it named a service category, whether it mentioned any organization, whether it made a recommendation of a specific firm, and whether it showed a source link.
Finding 1
AI search names the category almost every time
The buyer leaves knowing what kind of help to look for.
79 of 80 answers named a type of service. 56 of 80 named managed IT or a managed service provider explicitly. The result held in every product: ChatGPT, Perplexity and Google AI Overviews named a category in 20 of 20 answers, and Gemini in 19 of 20.
It held at every stage too. Even questions that only described a problem, like recurring outages or a phishing scare, came back with a type of service attached in 27 of 28 answers.
| Buyer stage | Answers | Named a service category | Recommended a firm | Showed a source link |
|---|---|---|---|---|
| Understanding the problem | 28 | 27/28 | 0/28 | 13/28 |
| Choosing a type of service | 28 | 28/28 | 0/28 | 17/28 |
| Choosing a provider | 24 | 24/24 | 0/24 | 12/24 |
| All answers | 80 | 79/80 | 0/80 | 42/80 |
Finding 2
No answer picked a firm for the buyer
Even when the question was about choosing a provider, though none of those questions asked for names.
None of the 80 answers explicitly recommended a specific firm. That includes all 24 answers to questions about choosing a provider. 10 of those 24 answers did mention organizations, such as software platforms, tools, publishers and MSP websites shown as source links, but 0 of 24 recommended a firm.
Organizations did appear. 35 of 80 answers named at least one, but that count includes public bodies, publishers and software platforms such as Microsoft, not only providers. A name in an answer is not a recommendation, and no answer turned one into a pick.
The provider answers gave what the questions asked for: selection criteria and red flags, such as response times, security scope, references, data residency and sales pitch warning signs. They equipped the buyer for the shortlist without writing it.
Finding 3
Whether you see sources depends on the product
The same question can come back with a page of links or none.
42 of 80 answers showed at least one source link. The gap between products was wide: Perplexity showed sources on every answer, Gemini on none.
Figure 1. Source links by product
Across those 42 answers there were 424 visible links. 28% of them pointed to government sites, such as US and Canadian cybersecurity and privacy agencies. Some provider blog pages also appeared, but as sources of explanation, not as firms the answer told the buyer to hire. For a related look at sources, see who Google's AI Overviews cited on marketing questions.
Canadian and US wording made little difference. Both sets named a category in 24 of 24 answers and recommended a firm in 0 of 24.
What this covers
How far these results go
A snapshot, read for what it is.
- This is a snapshot of 80 answers from September 2026. AI answers change over time and can differ by account, setting and location, so the same question may come back differently today.
- The 20 questions follow one buying journey in one market: managed IT and cybersecurity. None asked AI search to name or rank providers; the six provider questions asked how to evaluate, check or compare them. "Best provider for..." questions, or other industries, may behave differently.
- Canada and US results reflect how the question was worded, not where the person asking was.
- The results describe what the answers said. They do not show why a product chose a source, or what a provider could do to be shown.
Key terms and facts
The short version
- Service category
- A type of help, such as managed IT, managed detection and response, or backup and recovery, as opposed to a named company.
- Mention
- An answer names an organization. That can be a provider, a publisher or a public body. A mention is not an endorsement.
- Recommendation
- An answer explicitly tells the reader to choose or contact a specific firm.
- Source link
- A visible web link the answer shows as where its information came from.
Key facts
- TEAMULATE STUDY (80 answers, September 2026 capture)79 of 80 answers named a service category; 56 of 80 named managed IT or an MSP.
- TEAMULATE STUDY (80 answers, September 2026 capture)0 of 80 answers explicitly recommended a specific firm, including all 24 provider-choice answers.
- TEAMULATE STUDY (80 answers, September 2026 capture)42 of 80 answers showed at least one source link (Perplexity 20/20, Google AI Overviews 15/20, ChatGPT 7/20, Gemini 0/20).
- TEAMULATE STUDY (424 visible links)28% of the visible links pointed to government sites.
Takeaways
What to do with this
If you are buying
Use AI search to name the problem and the type of help, then build the shortlist yourself from the criteria it gives you.
If you are comparing answers
Check whether the answer shows its sources. In this capture that ranged from every answer to none, depending on the product.
If you sell services
In this capture, AI search handed buyers a category and a checklist. If your buyers use it the same way, the conversation may start at "how do you meet these criteria", not "what do you do". If you are weighing how to cover that work in your own marketing, see what an AI marketing team does.
Questions and answers
Common questions
Does AI search recommend specific IT providers?
Not in this capture. Across 80 answers to 20 buyer questions in four AI search products, none explicitly recommended a specific firm, including the 24 answers to questions about choosing a provider. Those questions asked how to choose, not who to choose. Results can differ with other questions, settings or dates.
What did AI search tell buyers instead?
It named the type of help. 79 of 80 answers named a service category and 56 of 80 named managed IT or an MSP. Answers to provider questions focused on what to check, not which firm to pick.
Which AI search products showed their sources?
Perplexity showed source links on 20 of 20 answers, Google AI Overviews on 15 of 20, ChatGPT on 7 of 20 and Gemini on 0 of 20.
Where did the source links point?
There were 424 visible links across 42 answers. 28% of them pointed to government sites, such as cybersecurity and privacy agencies in the US and Canada.
Did questions worded for Canada and the US get different answers?
Not much. Canada-worded and US-worded questions both named a service category in 24 of 24 answers, and neither set recommended a firm (0 of 24 each). Source links appeared in 14 of 24 Canada-worded answers and 12 of 24 US-worded answers. The comparison reflects how the questions were worded, not where the person asking was.
What came back when buyers asked how to choose a provider?
The six provider questions asked how to evaluate, check or compare providers, not who to hire. All 24 answers named a service category, 10 of 24 mentioned organizations such as platforms, tools or publishers, and 0 of 24 recommended a firm. The answers gave selection criteria and red flags instead.
How was the study done?
We wrote 20 buyer questions about managed IT and cybersecurity, with no brand names, across three stages: understanding the problem (7 questions), choosing a type of service (7) and choosing a provider (6). We asked each one in ChatGPT, Perplexity, Gemini and Google AI Overviews in September 2026, giving 80 answers, and recorded whether each answer named a service category, mentioned an organization, recommended a firm or showed a source link.
Sources
Named sources
Teamulate AI search discovery study: 20 buyer questions, 80 answers, September 2026 capture. Products queried:
- ChatGPT: https://chatgpt.com/
- Perplexity: https://www.perplexity.ai/
- Gemini: https://gemini.google.com/
- Google Search (AI Overviews): https://www.google.com/
Full report
Download the research
Visitor report · coming soon
Full report
The questions, answer counts by product and buyer stage, and limits (PDF).