How to get a haircare product recommended by AI assistants

This is the most common own goal, and it is free to fix.

OpenAI documents three separate crawlers, and they do different jobs. OAI-SearchBot is the one used to surface websites in ChatGPT's search features, so it is the one that affects whether you can be cited. GPTBot crawls content that may be used to train generative models. ChatGPT-User acts when an individual user asks something that requires fetching a page, and OpenAI states it is not used to crawl the web automatically and does not determine search visibility.

The practical consequence: a brand that blocked GPTBot to keep its content out of model training has not blocked itself from ChatGPT search citations, as long as OAI-SearchBot is still allowed. Plenty of teams block all three by accident, usually through a hosting provider's one-click AI bot blocker, and then wonder why they never appear.

Go and read your own robots.txt today. Not your policy about robots.txt, the file itself. Check what your CDN or host is injecting into it, because managed rules can add blocks your own configuration never mentioned.

An assistant answering a question builds the answer out of passages it can lift. A sentence that only makes sense after reading three paragraphs of brand story is a sentence that cannot be lifted.

So write the checkable facts as plain, self-contained statements. Not our revolutionary formula transforms your curl journey, but this leave-in is formulated for high porosity type 4 hair and contains shea butter and glycerin. The second one can be quoted into an answer and still be true and useful on its own.

A few concrete habits help:

• Put the specific facts, such as the texture range, key ingredients, size, and price, in text rather than only in an image.
• Answer the questions buyers actually ask on the page itself, in the words they use.
• State who the product is not for. This is the single most quotable thing most brands never write, and it earns disproportionate trust.
• Keep one clear page per product rather than one page describing eight variants, because a page about eight things has no quotable sentence about any of them.

Structured data is how a page states its facts in a form a machine does not have to infer.

Schema.org's Product type is the vocabulary for this, and Google documents how product and review markup is used for product snippets and review snippets. Google's product snippet documentation also describes positiveNotes and negativeNotes properties for expressing the pros and cons of an editorial product review.

Two rules matter more than the syntax. First, the markup has to describe what is actually on the page; markup that claims things the visible page does not say is a policy violation, not a clever trick. Second, Google's guidance is explicit that you should not include fake or undisclosed incentivised reviews on the page or in the structured data. That applies to seeded reviews, gifted products, and paid placements, and it is the same principle the FTC applies to disclosure.

There is more detail on the feed and markup side in product data for AI.

Here is the uncomfortable part. When someone asks an assistant which curl cream to buy, the answer is usually assembled from pages that are not the brand's own site. Reviews, forum threads, comparison articles, buying guides, and retailer pages.

That is not a quirk. A brand's own site is a self-declaration. It proves what the company claims, never that the claim is true. An assistant summarising opinion has an obvious reason to lean on sources with no stake in the answer.

So the highest-value work is usually the least technical: get the product into the hands of people who write about it independently, be present where your category is actually discussed, and accept that those write-ups will not all be flattering. A brand mentioned in twenty independent places with mixed opinions is far more likely to be named in an answer than a brand with a beautiful site and no third-party footprint.

This is also, bluntly, why partnerships like Kurli's exist. A placement or a seeded review is a third-party page discussing your product, which is exactly the material an assistant has to work with.

Be suspicious of anyone who says otherwise.

No one can guarantee a citation. The systems are not public, they change, and results differ between users and phrasings.

No one can give you a reliable ranking position in an AI answer, because there is often no stable ranked list to hold a position in.

Anyone quoting you a precise share of AI traffic for your industry is guessing, unless they can name the measurement method. Attribution from assistants to sites is genuinely patchy, and a lot of it lands in analytics as direct traffic with no referrer.

What you can control is real but less exciting: crawler access, page clarity, accurate structured data, honest disclosure, and third-party presence. Do those and you are eligible. Eligibility is the honest ceiling of what anyone can sell you.

1. Open your own robots.txt and read it. Confirm OAI-SearchBot is not blocked, and check what your host or CDN injects into the file.
2. Pick your three best-selling products. On each page, find one sentence stating who the product is for, in plain checkable words. If it is not there, write it.
3. Add or fix Product structured data on those pages, and confirm it matches the visible content.
4. Write the who this is not for line. Publish it.
5. List every third-party page that discusses your product. If the list is short, that is your actual problem, and no amount of on-site work will fix it.
6. Check that any gifted or paid coverage of your products is disclosed. Undisclosed coverage is both a legal exposure and a quality signal working against you.