In mid-September, AI assistants barely mentioned us when asked which UK B2B tech PR agencies to consider. We changed a handful of things on our website, and two weeks later ChatGPT was naming us in 47 of its 83 answers to the questions in our UK B2B Tech PR AI Visibility Index. Nearly all of that came from assistants reading our own site, which was the most useful thing we learned.
What we changed on our website
The biggest gap was a page that answered the buyer's question directly. Assistants that search the web look for exactly that, and we didn't have one. We did this in the first week:
- Wrote a page setting out which UK B2B tech PR agencies are worth considering and what each is good at, with our competitors described fairly.
- Added seven sector pages and a page on what B2B tech PR costs.
- Changed how our name appears in search results to "Resonance PR", because "Resonance" on its own is too common a word.
- Fixed thirteen old links that led nowhere, and retired out-of-date copies of our pages that Google still had on file.
- Put the key facts of every case study at the top, and added five new client stories.
In the second week we published three free research reports from our AI visibility data, each with an ungated PDF, along with three shorter articles on questions people search for. We also rewrote our main pages and all 27 case studies in plainer language.
What happened when the assistants read it
They found the new page within days. In our first run the assistants cited resonancecrowd.com in two answers. A fortnight later they cited it in 203, with that one page in 116.
The answers changed with it. ChatGPT went from naming us in 4 answers to 47, and Perplexity from none to 21. Grok took longer, perhaps because its search index picks up new pages more slowly than the others do: it didn't name us at all in week 1, and in week 2 it named us in 68 of its 83 answers.
All of this took us from 57th of 76 agencies to first, although the top four are within the margin of error and ChatGPT's weighting in the index is what puts us ahead. Count every assistant equally and we're fourth.
What the headline number hid
Almost every answer that names us is reading our site. Count only the answers that drew on other sources and we're 58th, roughly where we started, while Hotwire and CCgroup lead that view because journalists and review sites write about them in places they don't control.
A position like ours depends on one page staying the best answer, and it tells an assistant only what we say about ourselves. We now track the two views separately.
We also created a loop we hadn't planned for. The index page was cited in 81 answers, which means some assistants are partly reading our own report of the ranking when they answer.
Some of the change had nothing to do with us. We ask three check questions every week without scoring them, to see whether the assistants themselves have shifted. ChatGPT's answers held and the other five changed.
What we're doing next: earned media
Our research on over 1.3 million AI citations shows where independent material comes from. When buyers are choosing a supplier, about three in four of the media citations in AI answers come from trade and specialist titles, and trade articles are still being cited when they're about nine months old. The full findings are in Earned media in the age of AI answers, with a shorter guide to the trade publications AI cites most.
We'll pitch commentary and data to the trade titles that cover marketing and communications, try to get onto the third-party lists buyers use to build shortlists, and ask clients to write reviews on the directories and comparison sites the assistants read. Mentions from other sources are what we most want to grow from here.
The method and all 86 questions are on the index page, and there's more on how we work in B2B tech PR. To see how the assistants describe your company and which sources they rely on, ask us for a free AI visibility snapshot.
