Which AI assistants do buyers actually use, and how do they differ?

A prospect tells you they "asked AI" for a shortlist. Which one they asked matters. The answer a buyer gets from ChatGPT can differ from the one Microsoft Copilot gives inside their company's Microsoft 365, and both can differ from the AI Overview at the top of Google. One might name you, one a competitor, and one nobody.

The shift is big enough to plan for. In G2's April 2026 survey of 1,076 software buyers, 51% said they now start their research in an AI chatbot, up from 29% a year earlier. ChatGPT alone passed one billion weekly users in August 2026, according to OpenAI.

You cannot court six assistants separately. This guide sets out who uses which, how they differ, and the one body of work that helps with all of them, voice assistants included.

#Which AI assistant do B2B buyers use most?

ChatGPT has the largest audience of any standalone assistant, and Google's AI answers reach the most people overall because they sit inside Google Search. Behind them come Microsoft Copilot, which many buyers have at work, and Gemini, then Perplexity and Claude.

No company publishes UK-only figures, and we know of no UK survey of which assistant B2B buyers use to shortlist. The best guide is each company's own announcements, which we check every quarter.

The units differ, so the rows cannot be compared directly, and Similarweb counts website visits only, leaving out apps and AI built into other products. Business buyers also behave differently from consumers: Forrester found in January 2026 that they were twice as likely to use ChatGPT and four times as likely to use Microsoft Copilot. So work in this order: ChatGPT and Google, then Copilot and Gemini, then Perplexity and Claude. The wider change in how buyers choose is in how AI is changing B2B supplier choice.

#How do the assistants differ?

They differ in where their web results come from, whether they show their sources, and how fast a change to your site can appear. Every assistant mixes what its model learned in training with what it fetches when the question is asked. Answers from a live search can reflect a new page or article within weeks; answers from the model's memory change only when it is retrained, which takes months.

Each company documents the crawler behind its search answers: OpenAI says a site must allow OAI-SearchBot to appear in ChatGPT search, and Anthropic and Perplexity name Claude-SearchBot and PerplexityBot. Blocking a training crawler such as GPTBot does not stop you being cited; blocking a search crawler does. Cloudflare changed its default to block AI crawlers in July 2025, so check that setting too. The detail is in how AI crawlers read your website.

#Which search engine does ChatGPT use for web results?

OpenAI says ChatGPT search uses third-party search providers, alongside its own crawler and content from partners, without naming the providers. Microsoft's Bing was reported as one of them when ChatGPT search launched in October 2024. So look at how your firm appears in Bing as well as Google: submit your site in Bing Webmaster Tools and claim your Bing Places listing. The same work helps with Copilot.

#How do Google AI Overviews and AI Mode choose their sources?

Both build answers from Google's index using what Google calls query fan-out: several related searches for one question, with results drawn from all of them. That is why ranking helps but no longer decides. Ahrefs found in March 2026 that only 38% of AI Overview citations came from pages ranking in the top ten, down from 76% in July 2025.

AI Overviews are the summary at the top of an ordinary results page, shown in the UK since August 2024. AI Mode is a separate conversational search mode, live in the UK since 28 July 2025, where a buyer keeps asking follow-up questions. For both, a page that answers the first question and the likely follow-ups gives Google more reasons to draw on it than a page that ranks for one phrase. We compare the two in what Google AI Mode is, and explain why cited pages still get fewer visits in why traffic falls when rankings hold.

#Does Microsoft Copilot recommend suppliers differently from ChatGPT?

It can, because Copilot works from Bing and sits inside the tools many buyers use all day. Microsoft says that when web search is on, Copilot sends a query to the Bing search service and uses the results to compose its answer (Microsoft Learn, updated August 2026).

Microsoft 365 Copilot passed 30 million paid business seats in July 2026, according to Microsoft's earnings release, so a buyer writing a board paper can ask for a shortlist without going near Google. The extra work is looking after Bing: allow Bingbot, check that Bing has indexed your service pages, and make your Bing Places listing match your website. In our experience, a firm that only watches Google can be named by ChatGPT and still missed by Copilot.

#How do Gemini and Claude differ?

Gemini draws on Google's index and Google's understanding of entities (the firms, people and places it recognises as distinct things), so it rewards the same housekeeping as Google Search. Claude uses its own search crawler and cites a different mix of sources.

For Gemini, a complete Google Business Profile, consistent name and address details, and structured data (a short block of code that labels your firm, services and people for software) help it read your facts correctly; Google says its AI features need no special markup, so treat it as accuracy, not a ranking trick. Gemini keeps reaching buyers in new places: Google began moving phone users from Google Assistant to Gemini in March 2025, and Gemini in Chrome reached UK desktop users in July 2026.

The assistants favour different sources. In Muck Rack's May 2026 analysis of more than 25 million links cited by ChatGPT, Claude and Gemini, ChatGPT's most cited domain was Wikipedia, Gemini's was Reddit and Claude's was PubMed Central, and earned media (coverage written by others) made up 84% of cited links. For every assistant, "award-winning" without naming the award gives it nothing to repeat; specific, checkable statements confirmed by other sources give it something to say.

#Is Perplexity worth optimising for in B2B?

Not as a separate project, but it is worth checking every month. Its audience is small, 0.9% of visits to AI chatbot websites in Similarweb's September 2026 figures, but it searches the web for every question and numbers its sources, which makes it the quickest place to see why a competitor was named. A Peec AI study reported by Search Engine Land in March 2026 found G2 prominent in Perplexity's answers to B2B questions, so a gap in your reviews often shows there first.

#Why do ChatGPT and Gemini recommend different companies?

Because they read different things: different indexes, different favourite sources, memory in one case and a live search in another, and answers that vary from run to run. The disagreement tells you where the gap is, so ask each question several times and count a result only when it repeats. In our experience these patterns hold more often than not:

ID-Pal, an identity verification firm we work with, fitted the fourth row: "AI could describe ID-Pal. It never recommended it." After pages were rebuilt around the questions compliance leads ask, the assistants moved one at a time: "Gemini now names ID-Pal for UK wealth managers. Claude names it for UK fund administrators, on a page rebuilt in August, and links NorthRow to ID-Pal in the same answer." The detail is in the ID-Pal case study.

#Will Siri, Alexa or Gemini recommend our firm to a buyer?

They may, because voice assistants are being rebuilt on the same kind of models. Apple and Google said in January 2026 that Apple's next foundation models will be based on Google's Gemini and will help power a more personalised Siri. Amazon opened early access to Alexa+ in the UK on 19 March 2026; it routes each request to a model including Amazon Nova and Anthropic's Claude.

A spoken answer has room for one name, perhaps two. If an assistant would put you third on a written shortlist, you may not be mentioned aloud at all, so voice favours the firm that is plainly the answer to a narrow question. In our experience few buyers of B2B services shortlist by voice yet; local and professional services, where assistants read business listings and reviews, are further ahead. Test it by asking Siri, Alexa+ and Gemini your clients' two main questions aloud. The preparation is the same as for text, plus a specialism that fits in one spoken sentence.

#How do we prepare for all of them at once?

Build one foundation. Every assistant is checking what you do, for whom, and whether the rest of the web agrees; how they cross-check is in why ChatGPT sounds unsure about a firm.

#Step 1: Let the crawlers in

Allow Googlebot, Bingbot, OAI-SearchBot, Claude-SearchBot and PerplexityBot in robots.txt, and keep key text in the page's HTML: the main AI crawlers do not run JavaScript.

#Step 2: Write your specialism in one sentence

"A full-service agency helping businesses with all their digital needs" cannot be spoken as a recommendation. For example, "Demand generation for industrial manufacturers with sales teams of ten to fifty" can. Put your sentence at the top of your homepage and in every listing.

#Step 3: Claim the listings each assistant reads

Complete your Google Business Profile (Google and Gemini), Bing Places (Copilot) and Apple Business Connect (Siri and Apple Maps), even with no walk-in clients, plus LinkedIn and the directories in your field, with identical details.

#Step 4: Earn reviews where your buyers look

Reviews matter for B2B firms. In G2's April 2026 survey, 45% of software buyers said citations of review sites were the most confidence-inspiring signal in an AI answer. Ask satisfied clients for reviews on Google and on G2, Clutch or Trustpilot, and keep them recent.

#Step 5: Answer buyer questions on your own pages

Build pages around the questions buyers ask, with the answer near the top and evidence beneath: named clients, figures, named authors. See building content around customer questions.

#What to do this month

1. Write a fixed list of 20 buyer questions in a buyer's words, including the follow-ups they ask after the first answer.
2. Check robots.txt and any Cloudflare setting for the search crawlers above, and confirm in Bing Webmaster Tools that your service pages are indexed.
3. Put each question to ChatGPT, Google (AI Overview and AI Mode), Copilot, Gemini, Claude and Perplexity in a clean session, several times, recording who is named and which sources are cited.
4. Use the pattern table to find your most common gap, and fix it first.
5. Claim or complete Google Business Profile, Bing Places and Apple Business Connect so they match your website, with your one-sentence specialism at the top.
6. Ask Siri, Alexa+ and Gemini the same two questions aloud, then repeat the whole check in four weeks.

#Frequently asked questions

#Do we need a separate strategy for each assistant?

No. Check each one, because a buyer might use any of them, but the foundation is shared. Once your website and the sources that mention you tell the same story, the assistants tend to converge. The only platform-specific work most firms need is Bing for Copilot and Google Business Profile for Gemini.

#Should we pay for advertising inside the assistants?

ChatGPT has shown labelled ads to UK users on its Free and Go plans since 11 August 2026 (OpenAI); paid plans such as Plus, Business and Enterprise show none. OpenAI says the ads do not change the answer itself, so being named in the answer is still earned. An ad puts your name beside a conversation, not in the recommendation. We compare the two in whether being named by ChatGPT is cheaper than Google Ads.

#Does our Google ranking still matter?

Yes. Google AI Overviews, AI Mode and Gemini start from Google's index, so ranking helps, but only 38% of AI Overview citations came from top-ten pages in Ahrefs' March 2026 data, because Google runs several related searches for one question. Copilot starts from Bing, and ChatGPT search has drawn on Bing. Keep search healthy and cover the follow-up questions too.

#How quickly do changes show up?

Assistants that search the web live (ChatGPT search, Perplexity, Copilot and Google's AI features) can reflect a new page or article within weeks. Answers from a model's own memory change only when it is retrained, which takes months. ID-Pal shows the pace: "UK buyer questions where AI names ID-Pal: none in May, two in August, four in September 2026."

#Are there other assistants we should watch?

Yes. Similarweb's September 2026 figures show DeepSeek at 3.4% and Grok at 2.4% of visits to AI chatbot websites, and assistants are being built into browsers and office software. Because they all read the same web, a firm the main assistants name consistently is well placed for the next one. If buyers in your sector let agents build shortlists, read what happens when an AI agent builds the shortlist.

To see which assistants name you today, and why the others do not, book a free Growth Review: 30 minutes with David Adams and three fixes in writing. Where the gap is what other sites say about you, Reputation That Wins is built to close it.

#Sources