Where they started
The product was a solid project-management tool with a marketing site that read fine to a human and badly to a machine. Feature pages buried the answer three scrolls down, comparisons were framed as sales copy rather than facts, and the author of every post was a nameless “Team.” When we ran their 120 most valuable buyer queries (“best PM tool for [use case]”, “[tool] vs [tool]”, “how to run sprints in a small team”) through the major AI assistants, they were quoted on almost none of them. Competitors with worse products but cleaner, more quotable pages were getting named instead. The traffic was healthy and flat; the site was invisible in exactly the place buyers were starting to ask their questions.
What we did
- Rebuilt pages into self-contained passages. Each section answers one question in full, up top, without needing the paragraph before it. This is what makes a page quotable rather than just crawlable, and it’s what moved the citations.
- Fixed the entity and author signals: real named authors with credentials, consistent product naming, an organization identity a machine can resolve. Assistants cite sources they can attribute; “Team” is not a source.
- Opened crawler access and shipped an llms.txt. The AI crawlers had been getting blocked at the edge, so nothing else would have mattered until they could read the site at all.
- Built 30 links to the pages worth quoting. The citation lift was mostly on-page, but AI answers lean on pages that already have some independent trust, so the links backed the restructure rather than replacing it.
We tracked the same 120 queries every month, so the growth is measured against a fixed set, not a moving one. The curve is gentle on purpose: citations arrived a few queries at a time as pages were re-crawled and re-evaluated. By month nine the site was quoted on 41 of the 120, and roughly 9% of new signups arrived from an AI answer rather than a classic search result. We report that as a real, early channel, not a finished one.