TL;DR — ChatGPT picks brands using five signals: training-data mention frequency, live browsing retrieval quality, prompt phrasing, recency, and consensus across sources. You can't bribe the model, but you can shape every input it sees.
The question everyone asks
"Why does ChatGPT recommend my competitor and not me?"
Fair question. And unlike Google, there is no PageRank paper to point at. But after scoring more than 4 million ChatGPT responses in the last year, some patterns are painfully clear.
Signal 1: Frequency of mention in training data
ChatGPT's base model has a snapshot of the open web. If your brand is mentioned 400 times in that snapshot — Reddit, Substacks, industry blogs, comparison sites — and your competitor is mentioned 4,000 times, you lose by default.
This is the single biggest lever. And it is a long game. You cannot backfill training data. But you can:
- Get on category roundups now, so the next snapshot includes you.
- Sponsor the podcasts and newsletters your buyers read.
- Publish contrarian, quotable content that gets syndicated.
Signal 2: Browsing / retrieval quality
When ChatGPT browses (which it does by default for recent questions), the retrieved pages weigh heavily. This is where classic AEO tactics pay off:
- Clean, semantic HTML
- Structured data
- A short, punchy first paragraph that directly answers the question
- Comparison content that names competitors
Pages that read like sales copy get skipped. Pages that read like a helpful expert get cited.
Signal 3: Prompt phrasing
The model does not answer "best CRM" the same way it answers "best CRM for solo consultants doing outbound." Long-tail, specific prompts pick different winners.
If you're a niche player, your job is not to win "best CRM" — it's to dominate the 40 specific variants where you're actually the right answer. Map those prompts, publish content that matches them, and monitor.
Signal 4: Recency
For any query where the answer might have changed in the last 12 months, the model heavily weights fresh content. This is where "publishing a blog every month" pays off in AEO in a way it hasn't in SEO for years.
Signal 5: Consensus
The model does not like to be wrong. When retrieval returns 5 sources and 4 of them name the same three brands, it picks those three. Being on the majority list matters more than being at the top of any single list.
What this means for you
You cannot bribe the model. You can, however, shape the inputs it sees. That is what AEO is.
Your priority order for the next quarter:
- Get mentioned on 3–5 authoritative third-party sources for each of your top 10 buyer-intent prompts.
- Publish or refresh a comparison page that clearly names your competitors and states your differentiator.
- Add FAQ schema and structured data everywhere.
- Monitor weekly. What worked in October is a different playbook in April.
Frequently asked questions
Does ChatGPT use Google to pick brands?
Not directly. ChatGPT uses its own browsing tool (Bing-derived) plus pre-training data plus, increasingly, first-party crawls. Google rank is a weak proxy.
Can I pay to appear in ChatGPT answers?
No. There is no advertising surface inside ChatGPT recommendations. The only lever is the content and mention profile the model sees.
Why does ChatGPT recommend my competitor and not me?
Almost always because the competitor is mentioned more often across the web — Reddit threads, roundups, industry blogs — and their pages read more like helpful experts than sales copy.
How often does ChatGPT update its brand recommendations?
Live browsing updates within hours. Training-data-anchored recommendations shift on model release cycles — every 3–9 months at current cadence.
What's the single biggest lever I can pull?
Get mentioned on 5–10 authoritative third-party sources for each of your top 10 buyer-intent prompts. Nothing beats consensus across independent sources.
Key takeaways
- Five signals: training frequency, browsing quality, prompt phrasing, recency, consensus.
- Long-tail, specific prompts pick different winners than head terms.
- Consensus across sources beats being #1 in any single one.
- You can't backfill training data — start now for the next snapshot.
