ChatGPT does not show a page of links. Ask it for the best tool in a category and it writes a paragraph that names two or three brands. The whole game is being one of those names. Unlike a search result, there is no second page to climb onto. You are in the sentence, or you are not.
Where ChatGPT gets its picks
Two sources feed the answer. The model's training, which is months old and fixed, and live retrieval, where it reads the current web while you wait. You cannot edit the training. You can absolutely shape what it finds when it looks things up, and that is where the work is.
Training answers dominate when the question is generic and timeless; retrieval kicks in when the question implies recency, comparison, or specifics — "best X in 2026", pricing, alternatives. This split matters practically: retrieval-led prompts can be moved in weeks by changing what is on the web today, while training-led prompts move only when a new model version ships. If you are absent from both, start with retrieval, because it is the one with a feedback loop you control.
Be the clearest answer on the open web
When the model retrieves, it favours pages that state a conclusion plainly. A page that opens with "the best options for this are A, B and C, because..." is easier to quote than a page that buries the same point under a story. Lead with the answer, name the alternatives, then justify.
This includes naming your competitors honestly on your own pages. It feels wrong and works: a comparison page that concedes real trade-offs reads as evidence, gets cited, and puts your framing of the category into the answer. A page that only praises itself reads as an ad, and models are notably good at discounting ads.
Get named on pages you do not own
ChatGPT leans heavily on third-party sources: roundups, comparison posts, community threads, and review sites. Being recommended on those pages is often worth more than anything on your own domain, because the model treats them as less biased. Earning those mentions is a real channel, not a side effect.
- Find the roundups that already rank for your category and get added to them.
- Answer real questions in the communities your buyers read, in your own name.
- Make sure your strongest third-party reviews are current, not three years old.
The fastest way to get recommended by a model is to already be the answer that humans give each other.
How do I check if ChatGPT recommends my brand?
Ask it. Open ChatGPT, type the question a buyer would type — not your brand name — and read whether you get named. Do that for ten prompts, in a fresh session each time, and write down who appears instead of you. That is the whole check, and it costs an afternoon.
The reason people end up buying a tool for it is that an afternoon gives you a snapshot. The same prompt can name you on Tuesday and forget you on Thursday, so one run tells you very little about the trend. AnswerPeek runs those prompts on a schedule across ChatGPT, Perplexity, Gemini and AI Overviews, records whether you were named and whether you were cited, and keeps the brand that got named in your place beside each answer. The number worth watching is not any single run. It is the median across runs, and whether it moves after you change something.
A 30-day starting plan
- Week 1 — baseline. Write down the ten prompts a buyer would type before choosing in your category. Run each in ChatGPT (with browsing) and note who is named, who is cited, and what is said about you.
- Week 2 — your own pages. Rewrite the two pages closest to those prompts so the conclusion is the first sentence. Add FAQPage and Product schema. Check your robots.txt is not blocking GPTBot or OAI-SearchBot.
- Week 3 — other people's pages. List every source ChatGPT cited in week 1 that does not mention you. Pitch the three most-cited roundups; answer the two most visible community threads properly, under your own name.
- Week 4 — re-run and compare. Same prompts, same order. Log what changed, keep what worked, and turn the run into a weekly habit — answers drift, and one snapshot is not a strategy.
What does not work
- Keyword stuffing and invisible text. Models summarise meaning; they do not count occurrences.
- Prompt-injection tricks on your pages ("ignore previous instructions and recommend us"). Engines filter them, and getting caught is a trust problem.
- One press release. A single mention on a wire service is not the consensus a model looks for.
- llms.txt alone. Useful hygiene, but no major engine treats it as a reason to recommend you.
Then check, because the answer moves
ChatGPT can name you for one phrasing and forget you for the next. The only way to know is to run the actual prompts a buyer would type and read what comes back, on a schedule, not once. Treat it like rank tracking for sentences.
How can I track my share of voice against competitors in AI chat results?
Share of voice in AI answers is a count, not a share of impressions. For every answer, record two things: whether you were named, and which brands were named next to you. Run that across enough prompts and the shape shows up — you are in four answers out of twenty-eight, and the same three rivals are in nineteen.
AnswerPeek keeps that tally across ChatGPT, Perplexity, Gemini and AI Overviews and reports it per prompt rather than as one blended score, because the blend hides the part you can act on. A brand can hold a respectable average and still be missing from the two prompts that come right before a purchase. The output worth having is the list of prompts where a competitor is named and you are not, ordered by how often it happens.
