We’ve spent the last three months running a lot of branded queries — 4,200 of them, across GPT-4o, Claude 3.5 Sonnet, and Gemini 1.5 Pro — and watching what comes back. The same question, asked five different ways, ten times each, on each model. We logged every brand named, every URL cited, every claim about pricing or features.
The headline is boring: the models are not picking randomly. There is a recognisable signal, and it’s the same one across all three.
The signal
Brands that appear consistently share three traits:
- A clean knowledge graph. Wikipedia entry, schema.org Organization markup on the home page, consistent name across press, LinkedIn, Crunchbase. The models rely on these to disambiguate (is “Stripe” the payments company or the lighting company?) — get it wrong and you’re often left out of the answer entirely.
- High-volume, recent third-party coverage. Not first-party content on your own blog. We’re talking Reddit threads, comparison roundups on independent sites, YouTube reviews, podcast mentions. Sources the model can corroborate against itself.
- Pricing and feature accuracy that hasn’t drifted. When the model’s last training cutoff knew you charged $99 and you’ve since moved to $149, it will keep telling people about the $99 plan for months. That outdated mention becomes its truth about you.
Tags Research ChatGPT Citations