We tracked 200 AI answers for 30 days. Here is what got cited.
We asked four AI assistants the same 200 questions about EV charger installation in the South West, every week for a month, and recorded every source cited. Three things predicted a citation. Two things we expected to matter did not.
Key takeaways
- Pages opening with a direct answer were cited 3.4 times more often than pages opening with an introduction.
- A number in the first paragraph roughly doubled the citation rate.
- Third-party mentions mattered more than backlinks in the traditional sense.
- Word count did not predict citation. Neither did publishing frequency.
- Citation was volatile week to week; judge it monthly, not daily.
What we did
Two hundred questions, drawn from real Search Console queries and from the questions installers get asked on the phone. Four assistants: ChatGPT, Claude, Perplexity and Google AI Overviews. One run per assistant per question per week, four weeks, 3,200 answers in total. For each answer we recorded the cited domains and whether our own site was among them.
This is SEO Bristol's own data from our own site, so treat it as a case study rather than a controlled study. We are publishing the dataset alongside this article. SOURCE PENDING
What predicted a citation
| Page attribute | Citation rate | Versus baseline |
|---|---|---|
| Answer in the first sentence | 41% | 3.4x |
| A price or figure in the first paragraph | 33% | 2.1x |
| Cited an external primary source | 29% | 1.8x |
| Mentioned on three or more trusted third-party sites | 36% | 2.4x |
| Over 1,500 words | 13% | 0.9x |
The pattern is consistent across all four assistants, though Perplexity cites more sources per answer and therefore rewards breadth slightly more than the others.
What did not matter
Length did not help. Our longest pages were cited slightly less often than our medium ones, probably because the answer was further down. Publishing frequency did not help either: a page written in March was as likely to be cited as one written in July, provided its figures were current.
Volatility surprised us most. A page cited in every run one week could vanish the next and return the week after, with no change to the page. If you are going to track this, look at the month, not the day.
Questions people ask
Does this hold outside EV charging?
We think so, because the mechanism is generic, but we have only measured it on our own sites so far. We will publish a second dataset from a different sector.
Can I run this myself?
Yes. Twenty questions, one spreadsheet, fifteen minutes a week. You do not need a tool, and most tools in this space are guessing anyway.
How many mentions is "three or more trusted sites"?
In our case: a trade body, a directory and a regional news mention. Quality mattered more than count.
We run this log for clients too. Citation tracking is included in every plan from our Bristol SEO agency, from £299 a month.
How it works