What does AI actually look for?
Not clever copy. When an assistant decides which providers to name, it leans on information that is:
Specific
“24/7 monitoring from our own operations centre”, not “comprehensive solutions”.
Consistent
the same facts on your site, your listings and your partners' pages.
Confirmed by others
reviews, directories, analyst notes and press that say the same thing.
Readable
text on a web page, not locked inside a scanned brochure.
Current
dated, reviewed and free of products you no longer sell.
Where should you start?
With the questions, not the pages. Ask your sales team for the ten questions buyers ask before a shortlist, and write them in the buyer's words. “Who can onboard a mid-sized bank in under a month?” is a question. “Onboarding” is a keyword. Build the evidence around the questions.
What goes into an evidence base?
Five layers, in this order:
Fact pages
what you offer, who it's for, where you operate, how pricing works and which certifications you hold. Plain, dated and easy to quote.
Proof
case studies with specifics: the sector, the problem, what changed and a number where you can share one. “Helped a regional insurer” says nothing. “Cut claim response time from five days to two” says something.
Fit
who you're right for, and who you're not. It sounds counter-intuitive, but it's what makes AI confident enough to recommend you for the right questions.
Third-party confirmation
up-to-date directory listings, recent reviews, partner pages and coverage that repeat your core facts accurately.
Freshness
a review date on key pages, and a routine for retiring what's out of date.
What are the common mistakes?
- Important information that only exists in PDFs.
- Claims with no detail behind them.
- One service page trying to speak to every buyer.
- Different facts in different places, such as three versions of your office locations.
- Case studies with no outcome in them.
How do you check your evidence base?
Ask an AI assistant five of your buyers' questions. Where the answer shows sources, open them. Are they current? Do they say what the answer claims? Are they yours or someone else's? List every mismatch, give each one an owner, and fix the ones that would change a buyer's decision first.
Is this a guarantee?
No. Good evidence makes you easier to find, understand and recommend. It doesn't force any AI to name you. What it does is remove the most common, and most fixable, reasons you get left out.
Frequently asked questions
Do we need to write a lot more content?
Usually not. Most companies need fewer, clearer pages and better proof, not more blog posts.
Does structured data help?
It helps machines read your facts reliably. It's worth having, but it won't make up for missing or vague information.
How long before it makes a difference?
Fixes to your own pages are quick to make. How fast AI answers change varies by platform, which is why we measure over time rather than promise a date.