When a shopper asks ChatGPT "what's the best cold brew maker under $100?", the answer names a few specific products — and most stores are never among them. Getting recommended isn't luck, and it isn't something you can buy. It comes down to whether ChatGPT can find you, understand you, and trust you. This guide covers the concrete steps, in the order that actually matters.
How does ChatGPT decide what to recommend?
ChatGPT recommends products from two distinct sources: knowledge about brands baked into the model during training, and web pages it retrieves and cites live when it browses or searches. You need to work on both, because different questions trigger different behavior.
Trained knowledge is the model's long-term memory of the public web — your site, retailer listings, review roundups, forum threads, editorial coverage. If your brand is described consistently across many independent sources, the model can recommend you confidently even without browsing. This signal builds slowly and decays slowly.
Live retrieval and citations kick in when ChatGPT uses browsing or its search features, which is common for current, comparative questions ("best X in 2026," anything involving price or availability). Here it fetches pages, extracts facts, and cites sources — mechanics much closer to classic search. Your product pages compete to be fetched, understood, and quoted in real time.
The practical consequence: earned mentions across the web feed the first source, and crawlable, well-structured pages feed the second. Stores that only do one tend to be invisible for half the questions that matter. (For the bigger picture on why this is a distinct discipline, see our pillar guide, What Is Generative Engine Optimization?.)
Why are most stores invisible to ChatGPT?
Most stores are invisible because they fail one or more basic gates: AI crawlers are blocked, product data isn't machine-readable, or the brand simply doesn't appear anywhere the model reads. None of these show up in normal SEO dashboards, so stores rarely notice.
The common failure modes we see:
- AI crawlers blocked by accident. Blanket "block all bots" rules, aggressive CDN/WAF bot protection, or a robots.txt copied years ago quietly excludes GPTBot and friends.
- JavaScript-dependent product content. If price, specs, and description only appear after heavy client-side rendering, an AI crawler may see an empty shell.
- No structured data. Without Product JSON-LD, the model has to guess your price, availability, and product identity — and models prefer recommending what they're sure about.
- Vague product copy. Descriptions written as brand poetry ("elevate your ritual") give a model nothing to extract. It can't tell who the product is for or how it compares.
- Zero off-site footprint. If no review site, listicle, or community thread ever mentions you, there's no consensus for the model to echo.
Step 1: Allow AI crawlers in robots.txt
The first and cheapest fix is making sure OpenAI's crawlers — and AI crawlers generally — are allowed to read your product pages. Check your robots.txt for rules affecting user agents like GPTBot (model training), OAI-SearchBot (ChatGPT search), and ChatGPT-User (on-demand browsing), plus PerplexityBot and Google-Extended while you're there.
A minimal, permissive setup looks like:
User-agent: GPTBot
Allow: /
User-agent: OAI-SearchBot
Allow: /
User-agent: ChatGPT-User
Allow: /
Also verify your CDN or bot-management layer isn't blocking these agents at the network level — robots.txt can say "allow" while your WAF returns 403s. Fetch a product page with the crawler's user-agent string and confirm you get real HTML back.
Step 2: Add an llms.txt file
llms.txt is an emerging convention: a plain-markdown file at your domain root that tells AI systems what your site is and which pages matter most. It costs almost nothing to add and gives retrieval systems a curated map instead of leaving them to guess.
Keep it short and useful: a one-paragraph description of your store, then links to your top category pages, best-selling product pages, and key buying guides with a one-line description each. Think of it as a sitemap written for a reader in a hurry — because that's effectively what an LLM's retrieval step is.
Step 3: Ship Product JSON-LD on every product page
Structured data is how you replace a model's guesses with facts. Valid schema.org Product markup gives ChatGPT's retrieval unambiguous answers to the questions it needs settled before recommending you: what is it, what does it cost, is it available, what do buyers think.
At minimum, include:
name,description,brand,imageofferswithprice,priceCurrency, andavailabilityaggregateRatingand review markup where you have genuine reviewsgtin/skuidentifiers so the model can connect your listing to the wider product entity
Validate the markup, and keep it in sync with the visible page — mismatched prices between JSON-LD and the rendered page undermine exactly the confidence you're trying to build.
Step 4: Write product content a model can quote
ChatGPT recommends products it can describe accurately, so your product pages should hand it the description. The first paragraph of a product page should state plainly what the product is, who it's for, and its one or two defining strengths — in sentences that would survive being quoted out of context.
Beyond individual product pages, build comparison and buying-guide content around the questions your customers actually ask: "X vs Y," "best X for use case," "is X worth it." Use question-form headings and answer each question directly in the first sentence or two beneath it. These pages are disproportionately what AI engines cite, because they map one-to-one onto the questions users pose.
Honesty is a feature here, not a compliance cost. A comparison page that admits where a competitor wins reads as trustworthy to humans and models alike — and models are pattern-matching on trustworthiness at scale.
Step 5: Earn mentions where ChatGPT looks
Your own site can only take you so far, because "best of" answers lean on consensus across independent sources. The off-site work is unglamorous but decisive:
- Review sites and niche publications. Get your products into legitimate roundups and reviews in your category. One thorough independent review outweighs pages of self-description.
- Listicles and buying guides. When third-party "best X" articles exist for your category, being absent from all of them makes you absent from the consensus AI engines summarize.
- Reddit, forums, and communities — honestly. Community threads are heavily represented in what AI systems read and cite. Participate as a real vendor: answer questions, disclose who you are, be genuinely useful. Astroturfing gets detected by communities long before it helps you with any model, and burned trust is permanent.
There's no shortcut here. This is the GEO equivalent of link building, and like link building, the sustainable version is indistinguishable from building an actual reputation.
How do you know if it's working?
You measure by asking, repeatedly and systematically. Open fresh ChatGPT conversations, ask your category's core buying questions, and record whether your brand appears, in what position, and with what framing — then repeat on a schedule, because single spot-checks are noise.
Doing this manually across engines gets tedious fast, which is why the Rynex GEO Layer runs a fixed 9-run simulation panel — three model families (ChatGPT, Perplexity, Gemini) times three queries — and tracks your recommendation frequency and your competitors' over time. If you just want a baseline before committing to anything, the free GEO audit checks your crawlability, structured data, and actual AI mentions in under two minutes; our guide to reading the audit report explains each section.
Common mistakes to avoid
The fastest way to waste a quarter on this is to optimize the wrong layer first. Watch for these:
- Writing content while crawlers are blocked. No content strategy matters if GPTBot gets a 403. Always verify access first.
- Keyword-stuffing for a model. LLMs read meaning, not keyword density. Repetitive, over-optimized copy reads worse to a model than plain clear writing.
- Faking reviews or forum presence. Beyond the ethics, fabricated consensus is fragile — it contradicts the rest of the web, and models weigh the whole web.
- Checking once and concluding. One conversation where ChatGPT skips you means little; answers vary run to run. Trends across repeated runs are the only signal worth acting on.
- Ignoring the other engines. Perplexity and Gemini behave differently from ChatGPT. Visibility work overlaps heavily, so measure all three rather than tunnel-visioning on one.
Getting recommended by ChatGPT is not a trick — it's the compounding result of being crawlable, unambiguous, and genuinely present in your category's conversation. Start with the robots.txt check today; it takes five minutes and it's the gate everything else depends on.
FAQ
How does ChatGPT decide which products to recommend?
It combines what the model learned about brands during training with pages it retrieves and cites live when browsing or search is used. Consistent brand presence across the web feeds the first; crawlable, well-structured product pages feed the second.
Why doesn't ChatGPT recommend my store?
Usually because AI crawlers are blocked, product pages lack extractable facts and structured data, or your brand barely appears on the review sites, listicles, and forums the model draws on. Most stores fail on more than one of these at once.
Should I allow GPTBot in my robots.txt?
Yes, if you want visibility in ChatGPT. GPTBot and OAI-SearchBot need access to your product pages for OpenAI's models and search features to read and cite them. Blanket bot-blocking rules and CDN settings often block them unintentionally.
How long does it take to get recommended by ChatGPT?
Citation-driven visibility from browsing and search can move within weeks once your pages are crawlable and structured. Visibility rooted in the model's trained brand knowledge builds more slowly, over months of consistent presence across sources it reads.
How can I check if ChatGPT recommends my store?
Ask it your category's real buying questions in fresh conversations, several times, and record whether your brand appears. Rynex automates this with a 9-run panel across ChatGPT, Perplexity, and Gemini, and the free audit at rynex.io/geo-audit gives you a baseline in minutes.