How AI agents shortlist suppliers: what B2B buying agents read (2026)

AI agents shortlist suppliers by turning a buyer's brief into dozens of web searches. They read what comes back. Then they pull the facts into a comparison table and name the vendors whose facts they can find, check and see confirmed on other sites. Your website, your docs and the third-party pages that mention you are the raw material. If an agent can't read a fact, the fact doesn't exist for that shortlist, and most shortlists are settled before a buyer ever contacts you.

I'm Brenden House. I run Searchmaxxed, an SEO and AI SEO agency for enterprise, B2B SaaS, venture-backed, manufacturing and public-sector companies. On 27 October 2026 (AEST) I reread the help pages OpenAI, Microsoft, Google and Perplexity publish on how their tools search and fetch. Then I checked our own ChatGPT tests.

In summary

• Buyers already do this. Forrester found 94% of business buyers used AI in their buying process, and 61% use private AI tools their employer provides.
• Each agent searches its own way. ChatGPT uses its own search crawler, Copilot sends short queries to Bing, and Google's AI Mode fans one question out into many searches.
• Reading isn't citing. About three in four results it reads never make the answer, going by our own ChatGPT panel on 2 October 2026.
• Third-party pages carry the shortlist. Agents check your claims against review sites, lists and press, and they favour vendors whose facts agree everywhere.
• The fix is ordinary, careful work: readable pages for every buyer brief, facts that match across sources, and mentions on the sites agents trust.

Do B2B buyers use AI agents to research suppliers?

Yes. It's now the default first step. Forrester's 2026 buyer research found 94% of business buyers use AI in their buying process. AI now beats every other source. Forrester says "twice as many buyers" rated generative AI above any other source, "far outpacing vendor websites, product experts, and sales" (Forrester).

The tools sit inside the firewall. Forrester says 61% of business buyers use private AI tools their organisation provides. They're "four times as likely to use Microsoft Copilot as consumers" and twice as likely to use ChatGPT. For a supplier, that means Bing matters again, because Copilot searches through it.

Timing matters more than the tool. In 6sense's 2025 study of nearly 4,000 buyers, "Ninety-five percent of the time, the winning vendor is already on the Day One shortlist" (6sense). The same study found buyers lean on LLMs most in the middle of the journey. Their top use is comparing vendor offerings.

Gartner goes furthest. It forecasts that "By 2028, 90% of B2B buying will be AI agent intermediated", with over US$15 trillion of spend flowing through agents (Gartner). That's a forecast, not a measurement. The research shift is already here, though.

Which AI agents do buyers use, and how does each one research?

The main tools are ChatGPT, Microsoft Copilot, Gemini, Google's AI Mode and Perplexity. Each uses its own index and bot. That's why a vendor can lead in one tool and vanish in the next.

An AI agent is software that takes a goal, plans the steps and uses tools like search and a browser to return a finished result. Deep research is the mode that does this for a research question, then writes a cited report.

Some details from each maker's own pages:

• ChatGPT. OpenAI's help page says ChatGPT agent "is no longer available". Long, multi-step tasks moved to ChatGPT Work, and its cloud browser "can read web pages, click buttons, enter information into forms" (OpenAI). Deep research can "search the public web or specific sites, and use enabled ChatGPT apps", such as SharePoint (OpenAI).
• Copilot. Microsoft says Copilot "generates a short query based on your prompt and sends it to Bing" (Microsoft). Researcher, its deep research agent, has a web search toggle.
• Gemini. Google pitches Deep Research for due diligence, including "analyzing a company's products, funding history, team and competitive environment" (Google).
• Google AI Mode and AI Overviews. Google says both "may use a 'query fan-out' technique", meaning they issue "multiple related searches across subtopics and data sources" (Google Search Central).

One deep research setting deserves attention. A buyer can restrict research to sites they name, or tell it to "Prioritize these sites, but allow full-web search". If your category's review site or analyst is on that list, your profile there is the first thing the agent reads.

What happens, step by step, when an agent builds a shortlist?

An agent turns the brief into a plan and runs many searches. It reads what comes back, extracts the facts the brief asks for, then names and cites the vendors that fit. Vendors drop out at every step, mostly because the agent can't find or confirm something.

Here's a worked example. It's illustrative, built from how OpenAI and Google describe their tools, not a log of one real run.

A procurement lead at a 2,000-person Australian manufacturer opens deep research and types:

> Shortlist five supplier risk platforms that integrate with SAP Ariba, hold ISO 27001, keep data in Australia and suit a team of 20. Compare pricing model, implementation time and reviews.

1. Plan. ChatGPT "creates a proposed research plan" the buyer can edit before it starts. Each criterion becomes something to check.
2. Search. The agent breaks the brief into separate searches: the category, "best" lists, each vendor plus "SAP Ariba", "ISO 27001", "data residency" and "reviews".
3. Read. It opens the vendor pages those searches return: product, integrations, trust centre, pricing, case studies. Then review sites, lists and news.
4. Extract. It fills a table, one row per vendor and one column per criterion. Wherever it finds no fact, it writes "not stated" or leaves the vendor out.
5. Check. It hunts for the same claims elsewhere. A certification shown only in a gated PDF loses to one stated in text and repeated on a review profile.
6. Name. It returns five vendors with citations, and the buyer forwards the table to the other 12 people in the buying group.

Now picture vendor six. Its Ariba connector exists, but the only mention sits in a sales deck. Its data centre is in Sydney, but the website never says so. It might be the best product here. It still misses the list. Nothing the agent read said it fits, so as far as the shortlist goes, vendor six doesn't exist.

Our own testing shows the filter at work. We put 300 Australian agency-buyer questions to ChatGPT three times each, with search on. It pulled 17,870 search results into those 900 answers and kept 4,389 as sources. Being found is the easy part. About three in four results it reads never make the answer.

That panel is a one-day snapshot from 2 October 2026, and AI answers shift between runs.

What does an AI agent read on your website?

An agent reads the text it can get from your HTML, plus any structured data and markdown you serve. Headings, tables and plain statements come through best. The llms.txt proposal puts the problem well: turning a web page "back into clean text is difficult and imprecise". Content that loads only through JavaScript, text inside images and gated PDFs are the usual casualties.

Here's the set of pages an agent reaches for on a typical B2B brief, and what it needs to find there:

Two newer files help agents read faster:

• llms.txt is a markdown file at your site's root that summarises the site and links its key pages. The proposal behind it says "thousands of sites publish an llms.txt file" (llmstxt.org).
• Markdown versions of pages strip out navigation and scripts, and the same proposal suggests serving them at the page URL with .md added.

Keep them in proportion. Google's guidance for AI Overviews and AI Mode says to "apply the same foundational SEO best practices" (Google). It asks for no special file. Chrome's Lighthouse checks for llms.txt but marks a missing one "Not Applicable", since "providing the file is optional at the moment" (Chrome for Developers). Treat them as a cheap way to make your facts easier to fetch.

We run both on our own site. Searchmaxxed's llms.txt states what we do in two lines and links a markdown copy of every core page. Each copy names its canonical URL. Articles and service pages add a "Last updated" date, so it knows how fresh they are. Copy that pattern.

Schema helps only if it agrees. Google's AI features guide lists "Making sure your structured data matches the visible text on the page" among its best practices. Schema that claims something the page doesn't say gives an agent two facts to reconcile.

Where else do AI agents look, and why do third-party pages matter so much?

Agents check vendor claims against review sites, "best of" lists, directories, press coverage and forums. Those pages often decide the shortlist. Why? A few trusted sites turn up in a large share of answers.

We saw this in our ChatGPT panel. Across 900 ChatGPT answers about Australian agencies, a single third-party "best agencies" site was cited in 168 of them. The next three were a second listicle, Semrush's agency directory and Clutch. If you want to be recommended, the job is getting onto the handful of pages the model already trusts.

The engines differ, too. Perplexity cited a review directory (Semrush Agencies, Clutch, DesignRush and the like) in 36 of 44 agency answers. ChatGPT did it in 20 and Claude in 15. One agency's Clutch profile had six reviews and was cited 20 times.

That second panel ran on 29 September 2026. Treat both as snapshots of a moving target, not fixed rates.

Buyers behave the same way. Forrester says AI answers "often deliver incomplete or unreliable information, creating mistrust", so buyers seek "validation from trusted sources" (Forrester). Agent and human both look past you.

For a B2B supplier, the third-party pages agents read usually include these:

• Review platforms: G2, Capterra, Gartner Peer Insights, Clutch.
• "Best X for Y" lists that rank for your category searches.
• Trade press, including interviews, product news and customer stories that name you and say what you did for whom.
• Analyst and association pages.
• Partner marketplaces, such as the SAP Store or Salesforce AppExchange.
• Forums, Reddit included, where practitioners compare suppliers.

Keep your own profiles current. It's quick work, and it belongs to your team rather than an agency. Earning coverage on lists and in the trade press is a different, slower job, because editors and list owners decide when you appear. That one takes digital PR and links, which is where an agency like ours spends much of its effort.

What makes an agent name one supplier over another?

An agent names the suppliers whose facts match the brief, can be confirmed on other sites, and come from sources it trusts. Size helps less than clarity. A well-documented mid-sized vendor can beat a market leader whose pages say little.

Search rankings still feed all of this. ChatGPT, Copilot, Gemini and Perplexity all start from a search index. A page that ranks in Google and Bing for a category search is more likely to reach the agent's reading list.

Can you tell when AI agents visit your site?

Partly. Most AI crawlers and fetchers announce themselves by name in your server logs, so you can count them there. Analytics tools are far less reliable, because agent visits can look like ordinary traffic or like a bot attack.

Check robots.txt first. OpenAI says sites that block OAI-SearchBot "will not be shown in ChatGPT search answers" (OpenAI). Perplexity asks sites to allow PerplexityBot for the same reason (Perplexity). Perplexity's own docs explain how to allow its bots through Cloudflare and AWS firewalls, so check yours too.

User-triggered fetches follow different rules. For ChatGPT-User, OpenAI says "robots.txt rules may not apply" because a person started the request. Perplexity says Perplexity-User "generally ignores robots.txt rules". Google lists Google-Agent for "agents hosted on Google infrastructure" acting "upon user request" (Google).

Don't trust an "AI traffic" line in GA4 until someone checks it for bots, because agent visits and automated junk look alike in reports. Here's one enterprise B2B client's account. A client's GA4 showed AI-assistant visits jump from 354 to 6,050 in a month. Almost none of it was real. About 5,700 came in three one-hour bursts: same 1920x1080 desktop Chrome, all new users, no referrer, about 20 seconds each, landing on two pages. Real AI traffic was 385 sessions against 354 the month before, roughly flat.

The better measure is the answers themselves. Run the same buyer briefs through each tool every month and record which vendors each one names. Our guide to AI visibility tools compares the software that automates this.

How do you become the supplier AI agents shortlist?

Make every fact a buyer's brief might ask about readable, consistent and confirmed by sites other than yours. Then keep testing real briefs to see where you drop out, and fix those gaps first. Most of it is SEO.

Here's the checklist I'd run for a B2B supplier, in order:

1. Write down ten real buyer briefs. Take them from sales calls, RFPs and lost-deal notes, with the integrations, standards and constraints buyers state.
2. Run each brief through ChatGPT deep research, Copilot, Gemini and Perplexity. Record which vendors it names, which pages it cites and where you're missing.
3. Open the doors. Allow OAI-SearchBot, PerplexityBot, Bingbot and Googlebot in robots.txt and your firewall. Make sure Bing indexes your key pages.
4. Put key content in the HTML. Product facts, tables and specs shouldn't depend on JavaScript to appear.
5. Build pages for the briefs you want. Use case, industry, integration and comparison pages, each answering its question in the opening lines.
6. State the facts procurement checks. Certifications, data location, security standards, implementation time, support terms and how pricing works.
7. Write one fact sheet. Then make your site, LinkedIn, partner listings, review profiles and press boilerplate match it word for word.
8. Add llms.txt and markdown copies of core pages.
9. Earn third-party coverage on the lists, reviews and trade press agents cite for your category.
10. Measure monthly. Track the answers next to enquiries and pipeline.

Our AI search source readiness checklist goes deeper on steps 6 to 9. For ChatGPT specifically, see how to rank in ChatGPT.

Where does Searchmaxxed fit?

Searchmaxxed is an SEO and AI SEO agency for enterprise, B2B SaaS, venture-backed, manufacturing and public-sector companies. We do technical SEO, content, link building, digital PR, GEO, AEO and LLM SEO. The aim is simple: rank in Google and get named in ChatGPT, Copilot, Gemini, Perplexity and Google's AI answers when your buyers research.

Here, that means three things. We build and fix the pages agents read, in clean HTML with matching schema, llms.txt and markdown versions. We make your facts agree. And we earn the links and press coverage that put you on the lists agents trust when they build a shortlist for your category.

A page release is an approved new page or a substantial rebuild of one. Integration and comparison page families matter most here, because B2B briefs so often name an integration or a rival. Our AI search optimisation service sets out the full scope, and our enterprise SEO page covers large, multi-market sites.

Here's how we do it. We build on a website you own, our Agentic Website. AI agents pull your Search Console and GA4 data, read the Google results and check what AI answers say about you and your competitors. That feeds the Managed Search Loop, which shows which page, fix or link should win the most enquiries next. We build it, check the live page and measure what changed. Off your site, the Off-Page Source Layer builds the links, digital PR and mentions AI answers draw on.

Searchmaxxed's work has generated more than A$10 million in revenue through search for its clients.

Is Searchmaxxed right for you?

We suit marketing and growth teams at enterprise, B2B SaaS, venture-backed, manufacturing and public-sector companies who want the work built and shipped. If you want a dashboard of AI mentions and nothing more, we're the wrong call.

If you're in the right-hand column, a tool, a freelancer or your own team will serve you better.

Frequently asked questions

What should you do next?

Take three briefs your best customers would give an AI agent. Run them through ChatGPT deep research and Copilot this week, and note where you fall off. If you'd like us to fix what you find, let's talk pricing.