The most common question after an audit is some version of "so what should we write next". The honest answer is usually "nothing, until you fix five pages you already have".
Ecommerce brands have a strong reflex towards new content. It is what agencies sell and what content calendars are built around. But in AI answers, the pages that get pulled in are rarely the ones a brand is proudest of, and the gap between the two is worth understanding before you commission another post.
Five page types, and how they do
Product pages. Underrated, and for some assistants they are the main event. There is a meaningful split here: some assistants cite product pages heavily, others almost never, preferring editorial and third-party sources. The practical consequence is that your product page has to stand alone as a complete answer, because for a large share of buyers it is the only page of yours in the running.
Buying and comparison guides. Consistently cited, across assistants. This is the format assistants were effectively trained to summarise, and a good one written on your own site competes directly with the affiliate lists you cannot control.
How-to and care content. Quietly the best performer per hour invested. Seasoning a pan, storing whole spices, transitioning a dog onto new food, washing technical fabric. These questions get asked constantly, they have clear answers, and most brands have never written them down.
FAQ, shipping, returns and policy pages. Boring, unglamorous, and cited far more than anyone expects. Delivery times, duty responsibility, warranty terms and size guidance are exactly the facts an assistant needs and cannot safely infer.
Brand story and heritage pages. Beautiful. Expensive. Almost never cited.
Why the story pages lose
It is not a judgement on the writing. It is about what survives summarisation.
An assistant answering a question is looking for a fact it can state and attribute. Craft narrative, founder journey and mission copy carry feeling, and feeling does not compress into a recommendation. There is nothing in a heritage page that answers "does this ship to Dubai and who pays the customs".
We worked with a label whose artisan storytelling was genuinely lovely and had never once been cited. The pages that got cited were shipping, duties and fabric care. We kept the story pages, because buyers who arrive on the site do read them and they do convert. We just stopped expecting them to do a job they were never going to do.
That is the distinction worth holding. Some pages are for the buyer who is already with you. Others are for the machine deciding whether to send you one. They are not the same brief and they should not be written by the same instinct.
Audit your own site in an afternoon
This is the part you can do without any tools.
Run your prompt set and collect every URL that appears in a citation, yours and everyone else's. Put them in a spreadsheet with three columns: URL, whose site, page type.
Then count.
You will learn two things quickly. First, which of your page types earn citations, which is usually not the distribution you expected. Second, and more useful, which page types earn citations for your competitors. If every cited competitor URL is a comparison guide and you have none, that is your next quarter decided.
Most brands doing this for the first time find their single most-cited page is something nobody has touched in over a year. Usually a size guide or an FAQ.
What to do with the result
In roughly 80% of cases the answer is the same: rewrite five product pages before writing a sixth blog post.
Rewriting means stating, in text, on the page: who this is for, what it does better than the alternative, the specifications a buyer filters on, what it costs, how fast it arrives and what happens if they send it back. Not in a downloadable spec sheet. Not inside an image. In sentences.
Then, if the citation audit showed guides dominating your category, write two. Not twenty. Two good ones on the questions that came up most.
What does not work
Volume. Publishing thirty thin posts a month, whether written by a person or a machine, does not shift answers. We have watched brands do it for a quarter and land zero citations, while a competitor with one well-maintained care guide sat in answers the entire time.
Also: writing a new page instead of fixing the one that already ranks for the question. A second page competing with your own first page helps nobody, least of all an assistant trying to work out which of your two contradictory answers to believe.
Where to start
- Collect every cited URL from one round of prompts, yours and your competitors'.
- Sort by page type and count.
- Take your five best selling products and check each page states use case, specs, price, delivery and returns in words.
- Fix those five. Then decide whether you need anything new at all.