Getting Recommended by ChatGPT & Perplexity: GEO, Not Just SEO

How AI systems form answers about you

Put simply, generative systems combine two sources. First: the knowledge stored in the model from training - containing whatever was widely and consistently stated on the web at training time. Second: live research, where the system searches, reads sources, and cites them at runtime.

For individuals and mid-sized companies, the second source is usually decisive, because model knowledge is too thin. That's exactly why the question isn't "How do I get into the model?" but "What does a system find right now when it searches for my name - and is it unambiguous?"

Structured factual sources like Wikidata and Wikipedia carry disproportionate weight here, because they're machine-readable, documented, and easy to attribute.

Why the panel is the foundation

A Knowledge Panel is not itself an AI system, but it is visible proof that a clear, confirmed entity exists. The same signals that trigger the panel - a clear name, documented facts, independent sources - are the signals that guide an AI toward a clean answer.

Without that clarity, what we regularly see in audits happens: systems conflate people with the same name, invent plausible roles, or repeat outdated information. The self-test for this is in What ChatGPT Says About You.

Awareness, recommendation, fact-check

It's worth distinguishing three types of AI queries, because they reward different signals:

• Fact-check ("Who is X?"): Here, clarity matters. What's needed is correct, consistent core facts from documented sources.
• Awareness ("Who offers Y in Germany?"): Here, visibility in citable overviews, specialist articles, and directories that the system can find matters.
• Recommendation ("Who should I hire for Z?"): Here, third-party substance matters - editorial coverage, references, specialist publications. Self-description isn't enough.

Practical GEO steps

• Entity Home: a lasting page that factually describes your entity, including Person or Organization schema and a sameAs graph
• Consistent core facts: identical spelling, role, organization, and website across all sources
• Keep Wikidata clean: documented statements, external identifiers, no duplicates
• Create citable content: clearly structured articles with verifiable statements instead of marketing copy
• Editorial press: coverage that an AI can classify as an independent source
• Provide an llms.txt: a short, factual markdown summary of the key facts about your business
• Technical accessibility: no blocking of relevant crawlers, clean page structure, sensible headings
• Regularly check what different systems answer - and correct discrepancies at the source

What sets GEO apart from classic SEO

Classic SEO optimizes for rankings on search terms. GEO optimizes for appearing correctly within a generated answer - often without any click to your website at all.

That leads to different priorities: not keyword density, but factual clarity. Not as many pages as possible, but one clear source per statement. Not link volume, but citability. And above all: consistency, because contradictions in generative answers quickly lead to hallucinations.

What you cannot control

No one can guarantee that an AI system will give a particular recommendation. Answers vary by model, version, phrasing of the question, and point in time - the same question can produce two different results.

What can be influenced is the factual basis these systems draw on. That's exactly where we focus, and exactly why we don't promise placements in AI answers.

How to get started

Start with a stocktake: ask several systems the same question about your name and log errors, mix-ups, and missing information. That produces a surprisingly precise list of gaps.

After that comes the sequence from Creating a Google Knowledge Panel: Entity Home, structured data, Wikidata, databases, press. Ongoing monitoring falls under Monitoring & Maintenance. The free Brendit Entity-Score checks how you're represented across six AI systems.