← Guides

How to get recommended by ChatGPT and Perplexity

Updated 2026-08-03

Why ChatGPT recommendations are a system, not a hack

When an assistant like ChatGPT or Perplexity suggests a specific product or store, there's no secret button behind it. First the model has to find you, then read and understand your data, and finally trust you enough to say your name out loud. That's five layers of work, and they line up exactly with the categories in our scoring methodology: discovery, product data, logistics, hygiene and agent markdown.

The good news: across our rating of 63 MD/RO stores, the median score is just 38 out of 100. The bar is low, so systematic work moves you above average fast.

Step 1. Become visible to assistants

Assistants don't crawl with a single bot. ChatGPT search uses OAI-SearchBot; live user lookups come from ChatGPT-User; Perplexity uses PerplexityBot and Perplexity-User. If your robots.txt shuts them out, you're invisible — no matter how good the product is.

• Allow at least OAI-SearchBot and PerplexityBot in robots.txt.
• Keep an up-to-date sitemap.xml that lists every product page.
• Never block Googlebot or Bingbot — that kills the classic SEO signals assistants also lean on. Botmetria's firewall has this failsafe built in.

More in the robots.txt for GPTBot guide.

Step 2. Give data the model can't get wrong

Letting the bot in isn't enough — it has to read price and availability unambiguously. That's what JSON-LD schema.org/Product markup is for, with an offers block: price, priceCurrency, availability. Without it, the assistant guesses the price from page text and easily quotes an old one.

• Check that every product page carries valid JSON-LD Product.
• Keep availability set to InStock or OutOfStock, in sync with your stock.
• The "product data" category carries 30% of the score — the heaviest block, and for good reason.

See JSON-LD Product and availability.

Step 3. Serve clean text, not bloated HTML

Some AI agents prefer to read text/markdown over heavy HTML. Two tools help:

• llms.txt at the site root — a short, curated map of your store for models.
• Markdown versions of key pages for agents.

These cover the "discovery" and "agent markdown" categories (20% each). Start with the llms.txt for stores guide.

Step 4. Stop serving bots 403 and 406

The most common problem in our rating: product images answer bots with 403 or 406. The model sees a card with no image and recommends it less often. That's the "hygiene" category.

• Check that images open under bot user agents, not just in a browser.
• While you're at it, check response codes and page speed.

Walkthrough in the images 403/406 guide.

Step 5. Authority and fresh data

Indexability and markup are the entry ticket. To actually get *recommended*, you need trust signals:

• Mentions of your store on external sites, in reviews and niche catalogs.
• Real customer reviews.
• Fresh data: an assistant is happier quoting a current price and stock than something untouched for months.

Authority can't be bought in one move — it accumulates. But it won't fire without indexability and clean data, which is why the order of steps matters.

How long it takes — honestly

This is not a switch. Over two weeks, our AI Traffic Index logged more than a million AI-bot hits across a panel of three stores — yet live assistant visits were just 144 (90% from ChatGPT), and orders from AI were still zero. Bots are already reading stores heavily; human demand and purchases are only forming. The logic is simple: first you get re-crawled and re-indexed (that's weeks), and only then do you start surfacing in answers. Preparing systematically now is a bet on a channel that's growing.

Where to start in 15 minutes

1. Run your URL through the free audit and see which of the five categories is red (score ≤40).
2. Start with the heaviest and cheapest to fix — usually 403/406 images and JSON-LD.
3. Re-run the audit a couple of weeks later and check yourself against the rating.

FAQ

Related guides

• Why ChatGPT can't see your store: 7 reasons and fixes