How to use knowledge graphs to optimize for AI search
A small GEO operator claimed a +18.1% lift in active users after shipping a knowledge graph, plus referral traffic and key events above forecast. That doesn't prove causation. It does show something more useful for small SaaS teams: AI search may fail because the model knows the topic, but doesn't know who you are.
How does a knowledge graph help AI search know who you are?
A knowledge graph gives models a stable, machine-readable identity to attach your pages, product, founder, and profiles to. In Pixelmojo's write-up, they argue LLMs resolve entities, not just keywords, and they built 18 entities across 2 domains connected with JSON-LD @graph and sameAs links.
That claim lines up with a broader finding from Discovered Labs: page tweaks matter, but domain-level signals matter more. Their analysis of 2 million AI citations across 10,000 pages found that AI-perceived domain authority was 6 times more influential than the strongest page-level feature.
If you're a small software company, that explains a pattern you've probably seen:
- you publish solid pages on the right topic
- AI answers still cite better-known competitors
- your brand barely gets named, even when your content is relevant
The missing piece may not be "more content." It may be that the machine hasn't stitched your company into one clear entity yet.
More topical content won't fix entity confusion on its own
Publishing 50 more posts won't help much if your site still looks like scattered text. Discovered Labs found the strongest page-level lever was prompt-content alignment, with a standardised effect of +0.37, while the best on-page signal they measured, FAQ format, was only +0.07.
That matters because a lot of GEO advice collapses into "write more AI-friendly content." Useful, sure. But even the better data says content formatting signals are small compared with stronger alignment and domain-level trust.
Pixelmojo pushes the more technical version of that argument. Their point isn't that keywords stopped mattering. Their point is that models need something concrete to resolve: who the company is, what the product is, who the founder is, and how those objects relate across the web.
For a tiny SaaS team, that means your problem may be identity confusion, not topic coverage. You can rank for phrases, have decent docs, and still get ignored in AI answers because your brand signals don't connect cleanly.
A knowledge graph is mostly boring identity work, not magic markup
The practical work here is less glamorous than most GEO threads make it sound. Pixelmojo describes an entity foundation, a knowledge graph layer, machine-readable context, and multi-source authority. Strip out the packaging, and the job looks pretty simple.
You need to make the same company appear as the same company everywhere:
- one consistent company name
- one clear product name
- structured data that ties company, product, founder, and articles together
sameAslinks to the profiles and properties you actually control- repeated evidence across your site and other trusted places on the web
That last part matters. A knowledge graph on your own site is just you describing yourself. Models trust identity more when the same relationships show up in more than one place.
If you run a small SaaS, start with the pages that define you:
- homepage
- product page
- about page
- founder profile
- docs or key use-case pages
Then make sure the naming, schema, and linked profiles all say the same thing. Not close enough. The same thing.
The evidence on knowledge graph wins is early, so treat it like a test
The cleanest take here is skepticism. Pixelmojo openly says correlation isn't causation because they shipped other improvements at the same time. That's the right caveat.
But "entity work might help if AI search can't resolve your brand" is still a better hypothesis than vague advice to be helpful and wait. It's specific. You can test it.
Run it like conversion work:
- clean up your entity naming
- add or fix JSON-LD
- connect company, founder, and product with
sameAs - make external profiles match your site
- track whether AI answers start naming your brand more often
Discovered Labs is a useful brake on the hype here. Their data says alignment still matters a lot, and classic page features still help. So don't read this as a reason to ignore content. Read it as a reason not to assume content is the only problem.
If AI search knows your topic words but still skips your company, a knowledge graph is worth testing. Not because schema is magic. Because machines cite entities they can recognize, not brands they have to guess at.