For content and SEO leads

AI visibility for content and SEO leads

Optimising content for AI assistants shares most of its craft with good SEO, but the target is different: an answer that names, cites and recommends, not a place on a results page. Content teams should publish first for the buyer questions where a rival is chosen and they are absent, structure each page so the answer comes first, and measure the answers themselves. The three answers below cover the difference, the order of work and the shape of a page, followed by a checklist for page structure.

Talk to us See all services

A pale flamingo standing on bright yellow legs.

How is optimising content for AI assistants different from traditional SEO?

The craft overlaps more than it differs. Clear, specific, crawlable pages written for a real question help in both. The differences are in the target, the unit of work and the measure.

The target is an answer, not a ranking. A search results page lists many pages and the buyer chooses. An assistant writes one answer, cites a handful of sources and often puts a few companies forward, so the prize is being named, cited and recommended in that answer, on each assistant separately.

The unit is the question and the claim, not the keyword. Buyers ask whole questions with their sector, size and constraints in them, and the assistants decide for themselves whether to search at all. OpenAI and Anthropic each describe searching for some questions and answering others from what the model already holds.

Access has a new wrinkle. Both makers describe a search crawler that is separate from the crawler they use for training, so one broad block on AI crawlers can shut out search as well.

Independent sources matter more, because an answer may lean on a comparison article or a directory listing rather than on your site.

And the measure changes. There is no rank to read. You ask the same questions repeatedly and count how often you are named, recommended and cited, with an interval, because answers vary from run to run.

Which types of content should a B2B content team publish first if it wants AI assistants to cite its website?

Publish first where the evidence is missing and the commercial stake is highest, and find out where that is before writing anything. Read the answers to your buyer questions, note which are lost and which sources were cited, and sort the reasons into evidence gaps.

The first content type is a page for each priority question, titled in the buyer's own words, with the answer in its opening paragraph and the specific facts an answer would need. One page, one question.

The second is the fact pages: the systems you integrate with, the standards you meet, the sectors and company sizes you suit, each stated specifically and in the same words everywhere it appears. An integration described nowhere cannot be cited.

The third is a fair comparison or alternatives page, written for a buyer weighing options, naming other companies accurately and saying where each suits. A self-ranked list puts off both buyers and assistants.

The fourth is scope and pricing facts, where you publish them, because buyers ask cost questions and assistants answer them from pages that state it.

After those, work on independent sources: partner and directory listings kept current, and material a reviewer can quote accurately. Original data and method pages come later, once the basics are in.

Do not start with volume. Many generic articles close no gap.

How should B2B website pages be structured so AI assistants can quote them accurately?

Build each page so that a passage lifted out of it still makes sense and is still true.

Put the answer first. The title restates the buyer's question or a close variant, and the opening paragraph answers it in two or three sentences, with the detail underneath. Break the rest into plain headings, one sub-question each, so a quoted section stands alone.

Say who you are in full. Name the company and the product explicitly on the page, not only in the logo, and state the specific fact the buyer asked about in the same words you use elsewhere. Where two pages of yours describe a capability differently, the evidence for each is weaker.

Keep the text in the page. Content held in images, in scripts that load after the page, or behind tabs a crawler does not open may as well not exist for an assistant that reads the page.

Show a visible FAQ of real follow-up questions, answered in the text, and mark up only what is visible. Show dates: when the page was published, when it was updated, and when a changing fact was last checked.

Use plain markup that matches what the page shows. Neither OpenAI nor Anthropic asks for special schema. Use a table only where a comparison on stated attributes needs one.

Finally, let the assistants' search crawlers in, and check that your host does not block them.

A page structure checklist for AI answers

A question-shaped title: the buyer's question or a close variant of it, with the category words in the title and the heading.
The answer first: the opening paragraph answers in two or three sentences before anything else on the page.
A visible FAQ: real follow-up questions answered in text on the page, with FAQ markup built from that same visible text and nothing else.
Visible dates: published and updated dates, and a dated check beside any fact that changes, such as a price or a feature.
Plain markup: organisation, breadcrumb and article markup matched to what the page shows, with no special schema.
A table only where it serves: a captioned table for comparing options on stated attributes, and never a self-ranked list with the publisher first.
A caution on how to read this list: it describes features we saw in pages that assistants cited for our own buyer questions, and it is a guide to structure, not a promise that any page will be cited.

Questions content and SEO leads ask about AI visibility

Who is this page for?

It is for people who plan and write a B2B technology company's website content and own its search performance: heads of content, SEO leads, web managers and editors. If you run budget, finance or demand generation, there is a page for your role as well.

Do we need to stop doing SEO?

No. Pages that are specific, crawlable and written for a real question serve both. What changes is that you also measure the answers themselves, and put more weight on independent sources and on consistency of facts.

Do we need special markup or a special file?

The OpenAI and Anthropic pages we read ask for access, not for a special format. They describe separate crawlers for search and for training, so check which robots your robots.txt names and what your host blocks.

Does adding statistics and quotations make a page get cited?

There is research suggesting that adding sources, quotations and statistics can help in a benchmark, with effects that vary by subject. It tested rewritten passages, not live assistants, so read it as direction and not as a promise.

Should we publish FAQ pages for every question?

Publish a visible FAQ where real follow-up questions exist, and answer them in full. A page of thin questions added only for markup gives an assistant nothing to quote.

How do we know whether a page is working?

Ask the same buyer questions of each assistant before and after, several times each, and count how often you are named, recommended and cited. Compare like with like, and report flat results as plainly as the ones that moved.

Stand out.
Get chosen.

Create your account in under a minute. Start with your first project and the questions that matter.

Get started Talk to our team

White paper aeroplanes with one yellow plane leading.