About two months ago I started getting a recurring question from clients: "Why does our competitor keep showing up when people ask ChatGPT about our category, but we don't?"

I didn't have a good answer. So I built a simple tracking system and ran it for 60 days — 10 keywords, three AI platforms (ChatGPT, Perplexity, Google AI Overviews), two competing brands per keyword. Here's what the data actually showed.

The Setup

For each keyword, I asked the same question across all three platforms twice a week. I logged:

Nothing fancy — a spreadsheet and about 20 minutes every few days. No special tools.

Finding 1: Visibility varies wildly across platforms

The same brand could be well-represented on Perplexity and essentially invisible on ChatGPT — and vice versa. Platform behavior is not uniform, and this matters because your customers use different tools.

One brand I tracked had a 0% mention rate on Google AI Overviews for its core category keyword, while a smaller competitor with a weaker product appeared in 7 out of 10 queries. The difference wasn't brand size or product quality. It was content structure.

Finding 2: The AI's description is often wrong — and confidently so

In roughly 30% of cases where a brand was mentioned, the AI's description contained at least one material inaccuracy — wrong pricing tier, wrong target customer, wrong use case.

This is arguably worse than not being mentioned at all. A potential customer who gets an inaccurate summary may rule you out before they ever visit your site. And they have no way of knowing the summary is wrong.

Finding 3: Third-party citations are the real leverage point

The brands that appeared most consistently were almost never cited directly from their own website. The AI was pulling primarily from:

Your own homepage has surprisingly little direct influence on how AI systems describe you. The ecosystem of third-party content about your brand matters significantly more.

Finding 4: Structured content dramatically outperforms marketing language

This matched a framework I found in GEO: The New Enterprise Gateway in the AI Era (Huang Shuang, Li Jing, Wu Hao, Luo Wei, Yuntao Zou — City University of Macau, 2026), which the authors call "interpretation competition." The argument: AI doesn't reward better marketing, it rewards clearer, more structured, more verifiable information.

My data backed this up. Brands with product pages that included specific specs, certifications, pricing ranges, and comparison tables appeared in AI answers at roughly 3× the rate of brands with equivalent products but vague, marketing-heavy content.

The practical implication: a page that says "our enterprise-grade solution delivers ROI" gives the AI nothing to work with. A page that says "integrates with Salesforce, HubSpot, and SAP; setup takes 4–6 hours; pricing starts at $299/month for up to 10 users" gives the AI a lot to work with.

Finding 5: Negative content is sticky

If a brand had negative coverage — even old forum complaints — the AI would surface it alongside or instead of the brand's own positioning. Negative signals appear to be weighted heavily relative to their volume.

One brand in my data set had perhaps 15 negative forum threads out of hundreds of positive mentions online. Those 15 threads were appearing in roughly 40% of AI queries about the brand's reliability. Old negative content doesn't age out the way it might in traditional search rankings.

What Actually Moved the Needle

For the clients where I applied changes based on these findings, the actions that produced measurable shifts were:

  1. Rewriting product pages to include specific, structured information — not brand voice, just verifiable facts and specs
  2. Getting accurate coverage on 2–3 high-authority third-party review platforms
  3. Creating comparison content that directly addressed "[Brand] vs. [Competitor]" questions with fact-based tables
  4. Publishing case studies with real metrics — not "we helped Company X grow" but actual numbers with timelines

Caveats

This is a small sample. AI platform behavior also changes — I noticed noticeable shifts mid-experiment that seemed to correlate with model updates. Anyone claiming a stable, reliable GEO playbook is probably oversimplifying the current landscape.

That said, the directional findings feel robust. The brands that invested in structured, third-party-validated information consistently outperformed those that didn't, across all three platforms and across the full 60 days.

For a deeper theoretical framework behind why this is happening, the book GEO: The New Enterprise Gateway in the AI Era by Huang Shuang et al. is the most systematic treatment I've found — available on Amazon.