Quick answer
On 21 October 2025, Gartner predicted that "by 2028, 90% of B2B buying will be AI agent intermediated, pushing over $15 trillion of B2B spend through AI agent exchanges." Intermediated means an agent takes part somewhere in the purchase, not that agents make every decision. Gartner's own advice is to make products machine-readable. For a B2B seller that comes down to four jobs: put the facts buyers compare (pricing model, what's included, specs, security, integrations) in plain text on public pages; let the AI search bots read them; keep your profiles on review sites, marketplaces and procurement networks in line with your site; and keep reps ready to confirm what the agent found, because 69% of B2B buyers in a Gartner survey published in May 2026 prefer to validate AI-generated insights with a sales rep.
What Gartner predicted, word for word
The prediction is one of Gartner's top 10 strategic predictions for 2026 and beyond. It was announced in a press release on 21 October 2025 and presented by Gartner Fellow Daryl Plummer at Gartner IT Symposium/Xpo in Orlando. The exact line:
"By 2028, 90% of B2B buying will be AI agent intermediated, pushing over $15 trillion of B2B spend through AI agent exchanges."
The paragraph under it says verifiable operational data becomes "a currency", with digital trust frameworks and verifiability as "prerequisites for participation". It expects products built API-first and headless to gain a "competitive moat", and agents to "radically compress the sales cycle" for many business and technology purchases.
Gartner's conference highlights from the same day boil the advice down to making products machine-readable. Plummer's article of 14 November 2025 goes a step further: search engine optimization and pay-per-click "will give way to agent engine optimization", and products "will need to be machine-readable".
Three things the headline versions leave out:
- It's a forecast, not a survey. It was written for CIOs, and the release doesn't say how the $15 trillion was calculated.
- "AI agent exchanges" isn't defined. The release doesn't say what they are or who runs them.
- The word is intermediated. Many write-ups say "90% of B2B purchases will be handled by AI agents". That's a trade-press paraphrase, not Gartner's wording.
My work is outbound: cold email and LinkedIn for B2B companies, with 2000+ meetings booked for B2B clients. So I read this prediction from the seller's chair. When a buyer's agent looks you up, what does it find?
What "AI agent intermediated" looks like in 2026
Gartner's sales practice surveys real buyers, and its numbers are more down to earth than the prediction:
| Date | Gartner release | What it found or predicts |
|---|---|---|
| 25 Aug 2025 | Prediction | By 2030, 75% of B2B buyers will prefer sales experiences that prioritize human interaction over AI |
| 21 Oct 2025 | Prediction (above) | By 2028, 90% of B2B buying will be AI agent intermediated, over $15 trillion of spend |
| 9 Mar 2026 | Survey of 646 B2B buyers, Aug to Sep 2025 | 67% prefer a rep-free experience; 45% used AI during a recent purchase |
| 20 May 2026 | Survey of 645 B2B buyers, Aug to Sep 2025 | 45% used GenAI, mainly to gather information on vendors and products; 69% prefer to validate AI-generated insights with sales reps; 51% say they're more likely to get misleading information from GenAI, 49% from a rep |
So today the agent is mostly a research assistant. A buyer asks ChatGPT, Claude, Gemini or Perplexity to find and compare vendors, then checks the answer with a person. Gartner's May 2026 release describes the seller's role shifting from the main source of information to "a source of validation and confidence".
Inside larger buyers, agents already go further. SAP describes its Sourcing Assistant for SAP Ariba as creating sourcing events, recommending qualified suppliers and discovering new ones "from a network of millions", analysing bids and generating counteroffers. That's a procurement agent on the buyer's side of the table, and your supplier profile is its input.
Two kinds of buying agent, then: the research agent reads the open web, the procurement agent reads supplier networks and bid documents. A seller has to make sense to both.
What a buying agent can read on your site
Research agents reach your site through bots, and each AI company runs more than one. Block the wrong one and you drop out of AI answers:
| Company | Search bot | Fetch when a user asks | Training bot |
|---|---|---|---|
| OpenAI | OAI-SearchBot. Sites that opt out "will not be shown in ChatGPT search answers". | ChatGPT-User. OpenAI says robots.txt rules "may not apply". | GPTBot |
| Anthropic | Claude-SearchBot. Blocking it may reduce your visibility in search results. | Claude-User. Blocking it stops Claude fetching your pages when a user asks. | ClaudeBot |
| Perplexity | PerplexityBot, which Perplexity says isn't used to train models | Perplexity-User, which "generally ignores robots.txt rules" | None listed |
If you want to stay out of model training, block the training bots. If you want a buyer's agent to find you, keep the search bots allowed.
Then there's what a bot can see once it's in. Vercel's analysis of AI crawler traffic, published in December 2024, found that none of the major AI crawlers rendered JavaScript, including OpenAI's crawlers, ClaudeBot and PerplexityBot. Gemini, which uses Googlebot, did. It covered crawlers, not agents that drive a full browser, but it sets the floor: a price that appears only after a script runs, a calculator or a gated PDF may not exist for some of the bots that feed AI answers.
Google's guidance on AI Overviews and AI Mode, last updated 10 December 2025, is plainer than most "agent optimization" advice: no new machine-readable files or special schema needed. It asks for crawling allowed in robots.txt "and by any CDN or hosting infrastructure", important content as text and structured data that matches the visible text.
The CDN part catches people out. Cloudflare's changelog of 1 July 2026 lets every customer allow or block AI traffic by behaviour (search, agent and training), and since 15 September 2026 new domains block training and agent bots by default on pages that show ads. Check your CDN or firewall before you assume agents can read you.
A ten-minute test. Ask ChatGPT, Claude and Perplexity three questions your buyers ask: "What does [your company] cost?", "Does it integrate with [their CRM]?" and "How is it different from [your main competitor]?". Write down each answer and the sources it cites. Every wrong answer or "pricing isn't public" points at the page to fix first.
Pricing and spec pages an agent can compare
An agent comparing vendors needs the same facts as the buyer's comparison spreadsheet. Gartner's sales practice said as much in March 2026: structure content into modular, "agent-ready building blocks". Here's what that means page by page:
| Buyer's question | What a public page needs | Where it usually breaks |
|---|---|---|
| What does it cost? | A price or starting price, the pricing unit (per seat, per month, per volume), what's included and the minimum term | "Contact sales" only, or the price sits in an image or a script |
| Does it fit our stack? | Integrations listed in text, with what each one syncs | A wall of logos |
| Is it secure and compliant? | Certifications with dates, where data is stored, the DPA and the list of subprocessors | "Available on request" behind a form |
| How is it different from X? | A comparison page with facts: features, limits, pricing model | Adjectives, or a battlecard only reps can see |
| What happens after we sign? | Onboarding time, contract length, how to cancel | Nothing until the proposal |
If you can't publish a price, publish how the price works: the unit, what moves it up or down and what every plan includes. An agent can compare a pricing model. It can't compare "let's talk".
Write in plain sentences. "Plans start at [amount] per seat per month, billed annually, minimum five seats" is a line a buyer and an agent can both lift. "Flexible pricing that grows with you" isn't. And remember Gartner's condition, verifiability: for a mid-sized seller that means dated certifications and numbers that match across your site, your listings and your proposals.
Being listed where agents look
A research agent builds its answer from your site and from everything other sites say about you. Gartner's May 2026 survey found buyers used GenAI "primarily to gather information on vendors and products", so what other sites say about you matters as well.
- Search. Keep OAI-SearchBot, Claude-SearchBot and PerplexityBot allowed, and check that search engines index your pricing, integration and security pages.
- The sources the agents cite. Run the ten-minute test for your category, not only your brand: "best [category] for [segment] in Europe". The pages that get cited are your list: review sites, comparison articles, directories, partner marketplaces. Claim your profile where you can and make the facts match your site. If G2 shows up, here's what G2 badges actually measure.
- Procurement networks. If you sell to enterprises that buy through a procurement suite such as SAP Ariba, your supplier profile there is what a sourcing assistant reads first. Keep categories, regions, certifications and contacts current.
- Marketplaces your buyers use. If customers buy through a cloud marketplace or a partner's app directory, keep that listing as complete as your own site.
What doesn't change for outbound
Nothing in the prediction changes what earns a reply from a person. Gartner's own buyer research points the other way for the high-stakes part of a deal: 69% of buyers want to validate AI-generated insights with a rep, and reps remain the most important source when buyers pick a preferred supplier and finalise the purchase.
What changes is what happens after your email lands. A prospect who asks an assistant "what is this company and what does it cost?" gets the answer your public pages allow. If the pages say nothing, the agent fills the gap from someone else's page. Outbound gets your name in front of the buyer; your pages have to back up what the email promised.
If you put agents on your own selling side, the rule is the same: automate the work, not the thinking. AI owns the speed, you own the judgment. Before you start, read how to run an AI SDR pilot that stays out of Gartner's 40% of canceled agentic AI projects and what the agents inside your sales tools actually do. The wider tension between agents and buyers who want people is in what Gartner's AI predictions mean for outbound.
A 30-day plan for a B2B seller
- Week 1: test and unblock. Run the ten-minute test for your brand and your category. Check robots.txt and your CDN settings for the search bots above.
- Week 2: pricing in text. Put the unit, the starting price or pricing model, what's included and the minimum term in plain text. Remove the form in front of anything a buyer compares.
- Week 3: specs and trust. An integrations page in text. A security page with dated certifications and where data lives. A factual comparison page for each competitor you meet most.
- Week 4: listings and reps. Update the profiles the assistants cited. Brief reps on the questions buyers bring after checking with AI. Rerun the test.
Then run the test again every six months. Gartner's 2028 date is a forecast. What ChatGPT and Claude say about you today is a measurement.
Key takeaways
- Gartner predicted on 21 October 2025 that by 2028, 90% of B2B buying will be AI agent intermediated, pushing over $15 trillion of B2B spend through AI agent exchanges.
- Intermediated isn't "handled": today agents mostly research vendors for buyers, and 69% of buyers in Gartner's 2026 survey still prefer to validate what AI told them with a rep.
- Many AI crawlers don't run JavaScript, so prices, specs and security facts need to be plain text on public pages.
- Block training bots if you want to, but keep search bots like OAI-SearchBot and Claude-SearchBot allowed, and check your CDN settings.
- Make your profiles on review sites, marketplaces and procurement networks match your site, and test what assistants say about you every six months.
FAQ
What did Gartner predict about AI agents and B2B buying?
On 21 October 2025, Gartner predicted that by 2028, 90% of B2B buying will be AI agent intermediated, pushing over $15 trillion of B2B spend through AI agent exchanges. It was one of Gartner's top 10 strategic predictions for 2026 and beyond, presented at Gartner IT Symposium/Xpo.
Does Gartner say AI agents will replace B2B salespeople?
No. In Gartner's May 2026 survey of 645 B2B buyers, 69% preferred to validate AI-generated insights with sales reps, and Gartner predicts that by 2030, 75% of B2B buyers will prefer sales experiences that prioritize human interaction over AI. The rep's role shifts toward validation.
What does "AI agent intermediated" mean?
That an AI agent takes part in the purchase, for example by researching vendors, comparing offers or running a sourcing event. Gartner doesn't say agents will make 90% of buying decisions alone, and its release doesn't define the "AI agent exchanges" it mentions.
How do I make my B2B website readable for AI agents?
Put prices or your pricing model, what's included, integrations and security facts in plain text on public pages, not behind scripts, images or forms. Allow AI search bots in robots.txt and your CDN. Google says its AI features need no special files or schema.
Should I block AI crawlers like GPTBot and ClaudeBot?
Blocking GPTBot or ClaudeBot keeps your content out of model training but doesn't stop OAI-SearchBot or Claude-SearchBot. Those search bots decide whether you can appear in ChatGPT and Claude search answers, and OpenAI says sites that block OAI-SearchBot won't be shown in ChatGPT search answers.
