Guides · Updated October 31, 2026
Why ChatGPT doesn't recommend your product
ChatGPT recommends the products other people already vouch for. When it leaves yours out, it's for one of a few reasons, and its answers show which: who it recommended instead, the pages it read, whether it searched at all, and what it asked back. Here's how it picks, the reasons we see on our own products, and what the evidence says works.
How ChatGPT picks products
OpenAI documents its crawlers and its shopping results, but nothing about how an answer chooses, orders or tops its products. What we know comes from people measuring it. Put together, ChatGPT works like a careful researcher with a deadline:
- It decides whether to search. Without a search, it answers from what it learned in training, where a small or new product barely exists. Across all ChatGPT use, about a third of questions searched in early 2026 (Semrush, via Search Engine Land); nobody has measured the rate for buying questions.
- When it searches, it writes its own searches. One question becomes several, and software questions the most: about 10 per question in one study, many of them on vendors' own sites (Nectiv). It builds a shortlist from broad searches, then checks the shortlisted products' websites (Minuttia, 18,084 software questions).
- It reads other people's lists. For “best X” software questions, third-party “best X” lists are the pages it cites most: 43.8% of cited pages, twice the next kind, and the higher a product sits on a cited list, the likelier it's recommended (Ahrefs, 26,283 cited pages). G2 is the review site it cites most.
- Ranking on Google isn't enough. Only about 8% of the pages ChatGPT cites are in Google's or Bing's top 10 for the question (Ahrefs, 15,000 questions). Its own searches find other pages.
- When options look alike, the known name wins. In a lab test, well-known brands were recommended every time products were identical, and lost that edge as soon as a rival was clearly better on something concrete (Chu & Hou, 2026).
Shopping results (product cards, prices) are for physical goods. A SaaS product or a Shopify app has no card: it wins or loses in the text of the answer and the pages it cites.
One answer isn't a verdict
The same question almost never gets the same answer twice. Run 60 to 100 times, ChatGPT gave the same list of brands less than 1% of the time; how often each brand appeared was much steadier (SparkToro, 2,961 runs). Even two answers in the ChatGPT app disagree on the top pick about a quarter of the time, and the app and the OpenAI API, which some tools ask instead, share only 8% of the pages they cite (Recomma, 100 questions, raw data public).
So: ask the questions your buyers ask, in the ChatGPT app they use, several times, and count how often you're named, not where. Mentioned in 2 of 3 answers fits anywhere from about a 20% to a 95% chance; a question won or lost by most of its answers, read across many questions, is a measure. A screenshot isn't.
Eight reasons it leaves you out
For each question you're missing from, read the answers. Each reason leaves a sign.
1. It answered from memory
In the answer: no pages cited. It didn't search, so it named what it already knew.
What to do: nothing on your site changes this answer until a new model learns about you. Work on what it reads when it does search (reasons 3 to 6), and check whether your other questions search.
2. It answered for another buyer
In the answer: the products it names do a different job from yours.
On ours: SocialPhotos is a customer photo gallery for Shopify stores. Our questions said “customer photos”, which ChatGPT read as reviews apps, so it recommended Loox and Judge.me. They weren't beating us; it was answering another question.
What to do: find the words your buyers use for your category, use them on your site and listings, and ask your questions in those words.
3. It read pages that don't list you
In the answer: it cites roundups, comparisons, review sites or community threads, and you're on none of them.
On ours: for PodSeeker, one guide was cited in answer after answer, and it didn't list us.
What to do: list the pages cited most across your questions and ask their authors to include you (we wrote to that guide). This is the best-supported lever there is: in Ahrefs' own test, 94% of an established product's new mentions came from other people's pages, not its own (Ahrefs, 9,886 answers). Many cited lists sit on small blogs, so this is within reach. A G2 profile is close to an entry ticket for software; its review count barely moves where you're placed.
4. It read your pages and still didn't name you
In the answer: your site is in the citations, but your product isn't in the text.
On ours: PodSeeker had pages ChatGPT read without naming us. It's common: in Ahrefs' test, 43% of answers citing a page about a new brand didn't name the brand.
What to do: open those pages and check they answer the question asked, in plain words: what the product is, who it's for, what it costs, what it works with, and how it differs. ChatGPT checks shortlisted products on their own sites.
5. Your best page isn't on your website
In the answer: it cites your app-store listing, marketplace page or directory profile, not your site.
On ours: SocialPhotos' most-cited page was its Shopify App Store listing. Counting it as ours, SocialPhotos was cited in 9 of 16 questions, not 5.
What to do: treat those pages like your homepage. The Shopify App Store lets every AI crawler read listings, so its first lines are read whatever you do: make them say what the product is and who it's for.
6. Nothing sets you apart
In the answer: across your questions, it names the established products, and none of your pages are cited.
On ours: Askadia is new. Its free checkup found ChatGPT mentioned it in 1 of 10 questions and cited none of its pages. It recommended Peec AI, Otterly and Profound.
What to do: compete where you're clearly the best answer: a narrower buyer or need, stated concretely on your site and in the pages others write. A specific difference breaks the known name's lead; a generic claim doesn't.
7. The question is too broad
In the answer: ChatGPT asks back (“What are you trying to do?”) and offers options.
What to do: read the options. Each is a narrower question buyers ask, and you may win one of them. Track the one that's yours.
8. It can't read your site
In the answer: it names you from other pages, but never cites yours, on any question.
What to do: check these, in order:
- robots.txt allows OAI-SearchBot. It's the crawler behind ChatGPT search. GPTBot is only for training: blocking it doesn't remove you from answers, but a copied “block AI” list often blocks both (OpenAI's crawler docs). Changes take about a day.
- Your CDN or firewall lets it through. Cloudflare blocked AI crawlers by default on new sites from July 2025, and changed to separate search, training and agent settings in September 2026 (Cloudflare). Check its AI Crawl Control page, or any “block AI bots” rule.
- Your pages work without JavaScript. OpenAI's crawlers don't run it (Vercel): a page built in the browser, or a pricing table loaded later, is blank to them. Fetch the page with
curland look for your product's name and price. - Nothing important is hidden: no stray noindex, no feature page whose canonical points to the homepage, no pricing behind a login. A sitemap in Google Search Console and Bing Webmaster Tools costs nothing.
What works, and what doesn't
Nearly every study here comes from a company selling AI-visibility tools or services, us included. Prefer the ones with stated samples, and the few that changed something and measured it.
| Tactic | Verdict |
|---|---|
| Getting listed honestly on the “best X” pages in your niche | Works: the best-supported lever |
| A plain, readable site: what it is, who it's for, price, integrations, comparisons | Necessary: ChatGPT checks it |
| A narrow niche where you're clearly the best answer | Works: how small products beat known names |
| A G2 or Capterra profile | An entry ticket, not a ranking lever |
| Your own honest comparison or buyer's guide, authorship disclosed | Helps a little, least on ChatGPT |
| Paid placements on listicles | Unproven; must be disclosed |
| Many listicles ranking yourself first | Cited less and less; Google visibility drops reported |
| Planting posts and votes on Reddit | Don't: communities get banned and their citations vanish |
| Fake or bought reviews | Don't: illegal in the US (up to $53,088 per violation), and they barely move rank |
| Hidden text or instructions for AI | Don't: some models push the brand down instead; Microsoft treats hidden prompts for AI as an attack |
| llms.txt, schema markup, rewriting pages “for AI” | No measured effect |
Sources: self-ranking lists (Peec AI, 232,000 citations, Minuttia), Reddit (Otterly), reviews (the FTC's rule), hidden instructions (Microsoft), llms.txt (SE Ranking, 300,000 domains), rewriting for AI (C-SEO Bench; Google's own guide calls it “still SEO”).
Changes take time to show. Answers that search pick up a changed page in days to weeks, and some never do; answers from memory change only with a new model. Judge a change over a few weeks of answers, not the next morning.
Find yours
- Run the free checkup below: 10 questions your buyers ask, asked in the ChatGPT app, Google AI Overviews and AI Mode, every answer in full.
- For each question you're missing from, note who it recommended, the pages it cited (yours among them?), whether it searched, and whether it asked back.
- Match each to a reason above, and fix the one that comes up most (step by step, with our own numbers: how to rank in ChatGPT).
- Ask again every day for a few weeks before you call it fixed. That's what tracking in Askadia does, in ChatGPT and, if you choose, Google's AI Overviews and AI Mode, which often recommend someone else (why they differ).
With your agent, in Askadia (ours)
Give Claude Code, Cursor, Codex or any agent one line, Read https://askadia.ai/docs/api.md, and approve it once, or
add Askadia's MCP server (in Claude Code or Cursor, or as a connector in Claude.ai or ChatGPT). Then ask it to:
- List the questions you're missing from on each assistant, with who it recommended instead and the pages it cited.
- Read those answers word for word, with whether ChatGPT searched and what it asked back, and match each to a reason above.
- Check the questions themselves: one built from your own words is too easy, one a different buyer asks isn't yours. A question you lose isn't a bad question.
- After you fix a page, ask the questions it targets daily for a few weeks, then go back to Mon/Wed/Fri.
Included in the $29 plan, and reading is unlimited. It proposes changes until you tell it it can act, and the plan and its answers stay your click.
What nobody knows yet
The evidence is thinnest exactly where small products live. The biggest study left out sites below a moderate authority score; nobody has followed a product from unknown to recommended, studied app-store listings, or published a verified small-SaaS case with before and after numbers. We'll add what we learn on our own products here as it happens.
See which reason is yours
The free checkup asks ChatGPT, Google AI Overviews and AI Mode 10 questions your buyers ask, in about 2 minutes, with every answer in full.
100% free: no email, no sign-up, no credit card · Takes about 2 minutes