You’re Tracking Placements. You Should Be Tracking Citations.
Authored by: Hardik Parikh
Most experts can tell you how many articles they appeared in last quarter. Very few can tell you how often an AI assistant repeated what they said.
Those are different assets now, and the gap between them is where visibility is quietly being won and lost.
The distinction matters. A placement is a page with your name on it. A citation is that page being retrieved and quoted when a buyer asks an assistant a question. The first is easy to count. The second is what actually reaches the person deciding whether to shortlist you — and it does not follow automatically from the first.
The method
We measure it directly, and it takes about an hour a month.
Write down the questions a buyer genuinely asks before choosing in your category. Not keywords — questions. “What’s the best presentation software for agencies?” “How do I get an editable PowerPoint out of an AI tool?” We started with 69 of them.
Run each through grounded AI search and capture every source the model cites. Log the URLs. Repeat monthly.
That produced 575 distinct cited pages for us. That set — not the set that ranks in Google — is the shelf we’re actually competing on. Some pages that rank on page one are never cited. Some that rank nowhere are cited constantly. Until we ran this, we were optimising against the wrong list.
Three things it taught us
AI visibility isn’t one number. We’re cited consistently by Copilot and ChatGPT — roughly 6,700 citations a quarter across a single content cluster — and comparatively weak in Google’s AI Overviews. A single blended figure would have hidden that completely, and the fixes are different. Treat each assistant as its own channel.
The special files don’t do the work. We shipped llms.txt and per-page markdown twins, then retired both this July. The deciding evidence was accidental: our markdown layer served broken headers for weeks without us noticing, and citations continued at the same rate throughout. Crawlers were reading the rendered HTML. Those URLs now return 410, and the effort went into the page itself instead.
Retrieval doesn’t read rel attributes. A model summarising “the best tools for X” ingests text and repeats names. Whether a link is followed makes no difference to whether you appear in the answer. That inverts how most people score opportunities — an unlinked mention on a page assistants quote beats a followed link on a page they ignore.
What to do this month
Pick ten questions a buyer asks before shortlisting you. Run them through ChatGPT, Perplexity and Copilot. Write down which sources come back. Do it again in thirty days.
You’ll learn three things fast: who your real competitors are for the answer, which of your own pages are pulling weight, and whether the placements you earned last quarter are being read by anything other than people.
That list is the scoreboard. It won’t match your rank tracker, and it’s the one that decides whether a buyer ever hears your name.
Author bio: Hardik Parikh is the founder of Gixo.ai, an AI workspace for decks, proposals and business documents.