How a PR agency can show that coverage changed what AI says
You can show three things: that AI assistants cited the article, that your client was named more often on the same questions afterwards, and that competitors did not move the same way. You cannot show that the article caused it. An honest report says all four.
Set it up before the campaign, not after
The most common mistake is to look for evidence once the coverage is out. By then there is nothing to compare with. Before the campaign:
- Fix the questions. Choose the customer questions the coverage is meant to influence, and do not change them until the measurement is over. A new question set is a new experiment.
- Record a baseline. Ask those questions on the assistants that matter, several times a week, for about three weeks. Keep every answer.
- Name the competitors. You will need them as the yardstick.
What to record when the coverage lands
For every article: the exact address and the date it went live. That is all, but without the date nothing downstream works.
The three things to look at afterwards
| Question | How to answer it |
|---|---|
| Was the article used? | Did any answer cite that exact page, and did that answer name the client? This is observed, not estimated. |
| Did mentions move? | Compare the client’s rate before and after, on the same questions and the same assistants only. Three weeks each side is a reasonable window. |
| Did everyone move? | Check the competitors over the same window. If they all rose, the category got busier, and the article is not the story. |
What not to claim
- Not cause. Models are updated, competitors publish, news happens. “Mentions rose from 9% to 17% on these questions after the article, while competitors stayed flat” is a strong, true sentence. “The article raised AI visibility by 8 points” is not something anyone can know.
- Not a single screenshot. One answer that names the client is an anecdote. So is one that does not.
- Not page-level precision where there is none. Some assistants cite only the publication, not the page. Count those as context, not as a citation of your article.
What the client sees
One line per article: cited or not, in how many answers, and what happened to mentions on the questions it was meant to move, net of competitors. Clients trust a report more when it says what it cannot show.
NUR does this for agencies: each client’s coverage is tracked by address and date, matched against the pages the assistants cite, and reported next to the before-and-after on the same questions. How it works for agencies.