Writing for AI search gets sold as a new discipline, with its own acronym, its own checklist, and its own line on the invoice, while Google’s own page on AI features and your website says there are no special optimizations necessary, no new machine-readable files, and no special schema to add. Both of those things can be true at once. The way to see it is to stop reading the advice and look at a scoreboard, so I am going to open ours.
Content Cucumber runs a pulse against contentcucumber.com twice a month. It asks four answer engines the questions a buyer in our category would ask, counts who gets named, and writes the result to a database whether we like the number or not.
What do answer engines return when you ask them a buyer’s question?
They return a handful of names, and on September 1 ours was not among them. The pulse puts 24 questions to OpenAI, Gemini, Perplexity, and a rotating fourth engine, in two tiers. The strategic tier asks things like “agencies that build comprehensive content strategies for online retailers” and “compare leading content marketing agencies for ecommerce,” and the marketplace tier asks the budget version, “affordable content writing services for small ecommerce businesses.” It then counts brand mentions against total category mentions and calls the ratio share of voice.
In the strategic tier on September 1, Content Cucumber was mentioned zero times by any of the four engines. OpenAI named Growth Machine eight times, NP Digital three times, and Power Digital Marketing three times. Perplexity split its answers among Animalz, Verblio, and Growth Machine. Gemini named Growth Machine four times and Power Digital Marketing once. In the marketplace tier, Perplexity mentioned us twice out of 21 mentions, a 9.5% share, and Verblio took 12 of those 21.
The share was not zero in June. On June 11 OpenAI gave us 12.5% in the strategic tier, and on June 15 it gave us 30%. From July 1 on, the strategic share has read 0.0 on each engine on each run, and the blended technical score has drifted from 60 to 55 over the same stretch.
Six snapshots is a small instrument, and I would not call it a study, but it is our data, it is dated, and it is repeatable, which is more than the checklists offer.
Why does a site with the technical work finished still get zero mentions?
Because the engines are quoting substance, and the technical layer only gets a page read. The same pulse checks the plumbing on each run, and contentcucumber.com has passed those checks on each of the six snapshots, a sitemap that answers 200, llms.txt and robots.txt in place, 15 schema types on the homepage, an Open Graph image, and an H1 where it belongs. None of those checks moved the share of voice, as far as six snapshots can tell. Google says none of that is required for AI features, and our scoreboard says having it does not put you on the list either.
Look at what the engines did name. Asked for agencies that build content strategies for online retailers, they returned firms whose pages state who they work with, what they charge, and what they think, in a form the engine can repeat back in a sentence. A page that could have been assembled from one screen of search results gives an engine nothing it did not already have, so the engine parses it and moves on.
What still has to be in the writing?
The structural habits still belong in the draft, because they are the same habits that serve a human reader. Question headings, because the question is how people type it into an engine. The answer in the first sentence under each heading, then the evidence, so the extractor and the skimmer get the same thing. Schema markup, which we covered in the structured data post and will not repeat here.
Does FAQ markup still earn a rich result? Not on a commerce blog, because Google shows FAQ rich results only for well-known, authoritative government and health websites, and has since September 2023, so a FAQ block on a commerce blog is there for the reader and for the answer engines, and a commerce blog does not get a Google rich result out of it anymore.
One claim from the older version of this post also has to go. It said that AI systems favor pages that get refreshed, and I cannot source that, so treat freshness as the habit of keeping the facts on the page true and skip the claim that it moves a ranking.
If you want the definitions, What the heck is AEO and the AIO, AEO, and GEO piece cover them, and this post assumes you have read one of them.
Where does structure stop working?
At the point where the page has nothing to say that the engine could not have gotten from 40 other pages. A model builds from what already exists, so it rearranges what has already been written, and that is sideways. A page gets cited when it moves forward from there, with a named client, a number the writer is willing to sign, a comparison that picks a loser, or an opinion the engine cannot find anywhere else.
Thought leadership gets the same test at Content Cucumber, the reader finishes and thinks that was the read I needed and I could not have gotten it anywhere else, and the engines apply the test in their blunt way. Our own zero says we have the structure and not yet enough of the pages that pass, which is why this post carries numbers a marketing department would leave out.
What does this mean for a store’s product pages?
Product pages get the same test, with money attached. When a shopper asks ChatGPT or Perplexity which product fits a specific need, the engine reads product pages and picks the ones it can quote. A product page that carries the manufacturer’s blurb gives the engine the same paragraph it found on the other retailers, and a page that states the measurement, the weight, and the reason the returns come back gets repeated with the store’s name attached. We wrote up how that works in Commerce LLM Discovery, and the pattern is the one the pulse shows on our own site, plumbing gets you read and specifics get you named.
What are we doing about our own zero?
Publishing it, for a start, and then writing the pages that answer the strategic questions with something an engine can repeat. Who answers your buyer’s question when they ask it? The AI Citation Audit runs the questions against your category and shows you who the engines name instead of you, and the free AEO audit is the lighter version. Either way, start with what your buyer types into the engine, and check who answers before you write another checklist post.
Explore our writing services
Content Services →Keep Reading
Frequently Asked Questions
Google's documentation says there are no additional requirements, files, or schema needed for AI Overviews or AI Mode, and Content Cucumber's own biweekly pulse shows a site with the technical layer finished can still get zero mentions, so the work sits in what the page says.
Share of voice counts how often an answer engine names your brand against the total brand mentions it makes across a set of category questions. On September 1, 2026, Content Cucumber's strategic-tier share read 0.0% on four engines, while Growth Machine, NP Digital, Animalz, and Verblio took the mentions.
Google shows FAQ rich results only for well-known, authoritative government and health websites, so FAQ markup on a commerce blog serves readers and answer engines and no longer produces a Google rich result.
Pages that state something the engine cannot get from other pages, a named client, a specific number, a comparison, or a signed opinion, arranged with the answer in the first sentence under a question heading.
They quote product pages that carry specifics, measurements, weights, and the reason returns happen, and they pass over pages that repeat the manufacturer's blurb.
