Blog · Strategy

How AI assistants decide which B2B vendors to recommend

When a buyer asks an AI assistant for a shortlist, it isn't picking at random. Here are the signals that shape the answer, and the ones you control.

Ask an AI assistant "which MDR providers suit a mid-sized UK fintech?" and you'll get a tidy answer naming four or five companies, often with a sentence on why each fits. It feels authoritative. But where did that list come from, and why are some strong vendors missing from it?

No assistant publishes its ranking rules, and they change all the time. Even so, the way these systems work points to a fairly consistent set of signals. Understanding them is the difference between hoping you get mentioned and giving yourself the best possible chance.

Two ways an assistant "knows" about you

It helps to separate two sources of knowledge.

  • What the model learned in training. Large language models are trained on huge amounts of public text. If your company was discussed clearly and consistently across the web before that cut-off, the model has a sense of who you are. If you were barely mentioned, or described inconsistently, that picture is thin or wrong.
  • What it looks up in the moment. Many assistants now search the web while answering, then summarise and cite what they find. Perplexity does this by default. ChatGPT, Gemini, Copilot and Google's AI Overviews all do it for many queries. Here, the assistant behaves a bit like a very fast researcher skimming the results it trusts most.

The practical upshot: you need a strong, consistent footprint over time, and content that's easy to find, read and quote today.

The five signals you can influence

1. Entity clarity: does the AI know exactly what you are?

Assistants recommend things they can categorise confidently. If your homepage says you "empower resilient digital transformation", a model has little to work with. If it says you're "a managed detection and response provider for UK financial services firms with 200 to 2,000 staff", it has a category, an audience and a fit.

Check that your company name, category, audience and core proposition are stated plainly and consistently on your site, your LinkedIn page, review profiles, partner listings and press boilerplate. Inconsistency is one of the most common reasons we see assistants describe a company wrongly.

2. Content that answers the actual question

Buyers ask assistants specific, comparative questions: best options for a use case, alternatives to an incumbent, how to meet a regulation, what something costs. Assistants favour sources that answer those questions directly and in detail.

That means content most B2B sites avoid: honest comparison pages, clear "who we're for and who we're not for" pages, pricing approach explained, and pages tied to the frameworks buyers search for, such as DORA, NIS2, Consumer Duty or Cyber Essentials. Specifics beat adjectives every time.

3. Structure that's easy to extract

When an assistant is reading your page in a few hundred milliseconds, structure matters. Descriptive headings, short paragraphs that make one point each, lists and tables for comparisons, and a direct answer near the top of the page all make it easier to lift an accurate summary. Basic technical hygiene matters too: pages that load without heavy scripts, sensible structured data, and making sure you're not accidentally blocking the crawlers these assistants use.

4. Corroboration from sources the AI trusts

This is the signal most teams underestimate. An assistant is much more comfortable recommending you when independent sources say the same thing you do. In cybersecurity and fintech that typically means review platforms, analyst coverage, trade media, industry associations, partner directories, conference talks and community discussion.

Look at which sources an assistant cites when it answers questions in your category. Those are the publications and platforms worth being present in. It's often a short, very specific list, and it's rarely the one your PR plan was built around.

5. Freshness and consistency over time

Assistants that search the web tend to prefer recent, maintained content, particularly in fast-moving areas like threat intelligence or regulation. A comparison page last updated three years ago, or a product page that still mentions a feature you retired, works against you. So does a rebrand that left half the web using your old name.

What doesn't work

Because this space is new, there's a lot of noise. A few approaches to be wary of:

  • Volume for its own sake. Publishing dozens of thin, AI-generated articles adds little a model will cite, and can dilute your clarity.
  • Prompt "hacks". Hidden text or instructions aimed at AI crawlers are unreliable at best and a credibility risk at worst.
  • Guarantees. Anyone promising you a fixed position in ChatGPT is promising something they don't control. What you can do is measure, improve what's in your control, and track progress against competitors.

The short version

Be clear about what you are. Answer the questions buyers actually ask, specifically. Make it easy to read. Get independent sources saying the same thing. Keep it current.

None of this is magic, and most of it is good marketing that many B2B tech companies have let slide. The difference now is that an AI assistant is quietly scoring how well you've done it, every time a buyer asks.