What does a buyer actually type when they are deciding which CBD to order?
Almost never a product name. They type some version of “is CBD oil legit,” “how do I know a CBD brand is trustworthy,” “what should I look for on a CBD label,” or “best CBD for beginners.” Those are not shopping queries. They are trust queries, and they go into an assistant now as often as into a search box.
Whoever supplies the answer to a trust query gets the sale later. Most CBD brands are invisible at that moment, because everything they have published is designed to close a sale rather than to answer a question.
The question buyers ask before they ask about you
Sit with the buying sequence for a second, because it explains the whole strategy.
A person hears about CBD, becomes curious, and immediately becomes suspicious, which is the correct response to a category with a documented history of mislabeled potency. So the first questions are defensive. How do I know what is in it. Who tests it. Is this legal where I live. What does full spectrum mean and does it matter.
Only after those are settled does anyone ask which brand to buy. By then the assistant has already formed a picture of what a trustworthy brand looks like, built from whatever sources it found, and the brands that appear in the answer are the ones that were present in the earlier explanations.
That is the opening. AEO for CBD brands works by owning the defensive questions, because the brand that teaches someone how to evaluate the category gets evaluated favorably by the standard it just taught.
Your product pages are not competing for that. They were never in the running.
Why this is the hardest category to get cited in
Three forces stack against a CBD brand, and pretending otherwise leads to wasted budget.
The first is regulatory. The FDA has not approved CBD as a dietary supplement and has said the existing framework does not fit the category, while continuing to issue warning letters to companies making health claims. Assistants are trained to be careful around health topics generally and especially careful where a regulator has been active. Content that edges toward effects gets discounted or ignored.
The second is a credibility deficit earned by the category itself. Independent testing over the years has repeatedly found products whose actual cannabinoid content did not match the label. That history is in the training data. A model reading about CBD has absorbed a lot of material about mislabeling, which raises the bar for what counts as a trustworthy brand.
The third is structural sameness. Visit fifty CBD sites and you will find fifty versions of the same page: hemp-derived, third-party tested, non-GMO, made in the USA, a link to a COA PDF. None of it distinguishes anyone. A model cannot prefer a brand on the basis of language that every brand uses.
Taken together, these explain why the answers in this category skew toward large retailers and editorial roundups rather than toward brands. The path through is evidence that is specific enough to be quotable.
The Substantiation Trail
I use a simple structure for this called the Substantiation Trail. It has three links, and a brand that completes all three becomes citable in a way that competitors with the same claims are not.
The first link is the batch. Not a generic testing claim, but a page per batch with the lab’s name, the test date, the cannabinoid profile in numbers, and the contaminant screens performed, written out as text. Most brands bury this in a PDF or behind a QR code, which means the actual numbers exist nowhere a machine can read them. Put them in HTML.
The second link is the source. Where the hemp was grown, in which state, under which agricultural program, and what the extraction method was. This is public-facing supply chain detail, it is verifiable, and almost nobody publishes it with real specificity because it feels operational rather than promotional.
The third link is the independent mention. Somebody who is not you, saying your name in a context a model trusts. A retailer listing with real detail, a trade publication, a local business story, a comparison piece you did not write or pay for.
Each link on its own is weak. Together they produce something a model can assemble into a sentence: this brand publishes per-batch lab results from a named lab, sources from a specific program, and is mentioned by outlets that are not the brand. That sentence is what a citation is made of.
Build the pages that answer, not the pages that sell
Take the defensive questions from the first section and build one page per question. Not a blog post that meanders toward a product. A page whose title is the question and whose first two sentences are the answer.
How to read a CBD certificate of analysis. What full spectrum, broad spectrum and isolate actually mean for what is in the bottle. Is CBD legal in my state, with the actual variation spelled out. What the 2018 Farm Bill did and what the 0.3 percent THC threshold means in practice. Why two products labeled the same strength can differ. What a COA does not tell you.
Write them so that any two-sentence extract stands on its own. That is the practical test for whether a passage can be pulled into an answer. If the second sentence only makes sense after reading the first four paragraphs, rewrite it.
Then hold the line on what you will not publish. No dosing guidance. No condition-specific claims. No wellness language dressed up as education. Every one of those is both a regulatory risk and a reason for a model to downgrade your entire domain as a health source.
Restraint is what makes the rest of your content credible.
Third-party proof outranks your own COA page
Here is the uncomfortable ranking. When a model decides which CBD brands to name, the order of weight runs roughly like this: independent editorial coverage, major retailer listings, review platforms and forums, structured directory data, and last, your own website.
You cannot self-declare your way into an answer. Every brand in the category makes identical assurances on its own domain, and a model that weighted self-description heavily would produce useless answers.
So spend proportionally. If you are putting ninety percent of your budget into your own content and ten percent into being mentioned anywhere else, the ratio is backwards. Earned mentions do the heavy lifting.
The realistic targets for a small brand are trade and industry publications covering hemp and supplements, local business media in the state where you operate or manufacture, product comparison sites that accept submissions for independent testing, and retailer relationships where the retailer’s own page carries real detail about you.
One substantive third-party source that names your brand and describes your testing does more for AI visibility than another twenty product pages.
What to do in the next thirty days
Start by auditing the actual answers. Put fifteen buyer questions into ChatGPT, Perplexity, Gemini and Claude, and write down which brands get named and which sources get cited. The cited sources are your target list, because those are the places a model already trusts in your category.
In week two, convert your COAs from PDFs into readable pages with the numbers in text, one per batch, linked from the product they belong to.
In week three, write the first five question pages, starting with how to read a certificate of analysis, because it is the highest-intent defensive question in the category and it positions you as the brand that taught the buyer their standard.
In week four, pitch one trade or local outlet on a story about testing, labeling accuracy or supply chain, which are the angles a business desk will run and a product pitch is not.
Then wait, because none of this moves in under a quarter, and re-run the same fifteen questions at day ninety against the baseline you recorded on day one.
If you do one thing this week, open your most popular product page, find the certificate of analysis behind it, and publish those numbers as text on a page of their own.