Every week we ask six AI assistants the same 72 questions that buyers ask about PR software and AI visibility tools, and count which tools they name. The PR software AI index started on 13 October with Agentcy 74th of 77. Last week it was 23rd, named in 23 answers. This is week five, and it's the first week we went backwards.
The numbers
| Run | Assistants scored | Answers | Answers naming Agentcy | Agentcy's rank | Answers citing agentcy.com.pr |
|---|---|---|---|---|---|
| 13 Oct (baseline) | 5 | 357 | 1 | 74th of 77 | 1 |
| 20 Oct | 5 | 349 | 14 | 26th of 76 | 11 |
| 27 Oct | 6 | 420 | 24 | 21st of 72 | 24 |
| 3 Nov | 6 | 430 | 23 | 23rd of 76 | 28 |
| 10 Nov | 6 | 416 | 18 | 25th of 76 | 31 |
Agentcy was named in 18 of 416 answers and fell two places to 25th. By assistant, ChatGPT named it in 7 answers (9 last week), Google AI Overviews in 6 (5), Google AI Mode in 4 (4) and Grok in 1 (2). Gemini and Perplexity named it in none, after 2 and 1 last week.
A rate limit, and a change to how we count
Google AI Overviews answered only 47 of the 72 questions on the first pass, against 70 last week. The service we use to collect Google's results limited our requests part-way through the run. On that first pass alone, Agentcy was named in 14 of 407 answers and ranked 32nd.
Whenever an assistant fails on a question, we ask it again the same day. This week that retry recovered nine AI Overviews answers, and four of them named Agentcy. Until now we didn't count retry answers. That was the wrong rule: a collection fault on our side shouldn't decide a tool's rank, ours or anyone else's. From this week, where an assistant failed on a question and the same-day retry answered it, the retry's answer counts. An answer from the first pass is never replaced. Earlier weeks don't change, because their retries recovered nothing.
The fair comparison is the 414 question-and-assistant pairs answered in both weeks. On those, Agentcy went from 21 answers to 18. So part of the fall is the missing AI Overviews answers, and part of it is real.
Recommended, not just cited
Of the 18 answers naming Agentcy, 13 recommend it and 5 only use one of our pages as a source. Last week it was 16 and 7.
What we gained
- pr agency software. Google AI Mode lists Agentcy as "tailored specifically for modern B2B and tech PR agencies, offering per-client pricing, integrated journalist CRMs, and AI-visibility tracking". This is the broadest question we've been recommended on.
- We are a 12-person B2B tech PR agency paying for Cision. We want to cut cost and be better at AI. What should we move to? Google AI Overviews lists Agentcy under the heading "Best AI-Native B2B Software". It's the first time we've been named on a question about leaving a rival.
- Which PR tool tracks AI visibility as well as media coverage? AI Overviews names Agentcy for the first time on the question closest to what we do: it "integrates weekly AI assistant tracking (ChatGPT, Gemini, Perplexity) directly alongside its media monitoring, journalist CRM, and coverage reporting tools".
- Which PR software is priced per client for agencies? Google AI Mode joins Grok and AI Overviews.
- What is the best PR software for a one-person in-house comms team? and What is the cheapest way to monitor press coverage of my company? AI Overviews names Agentcy on both.
One of those needs a correction. On the cheapest way to monitor coverage, AI Overviews points to our free plan. The free plan reads your sector's news and holds your journalist CRM, but it doesn't find your coverage for you; media monitoring comes with Pro. Being named is not the same as being described correctly, and we check for both.
What we lost
- Gemini and Perplexity. Gemini named Agentcy on free PR software and on client logins last week, and on neither this week. Perplexity dropped its one answer, on UK B2B tech agencies.
- ChatGPT on per-client pricing. It had named Agentcy there in both of the previous two weeks.
- Grok on AI visibility and media coverage. Grok moved to a newer model this week, so a different model is answering. That question now goes to Muck Rack, PRCoverage and Meltwater.
- Google AI Mode on client logins and on signing up without a sales call. ChatGPT still puts Agentcy first on client logins.
Cited more, named less
agentcy.com.pr was cited in 31 answers, up from 28, which made it the sixth most-cited site in the run, behind muckrack.com (63), meltwater.com (58), prezly.com (52), cision.com and pr.co. All six assistants cited it again.
But 15 of those 31 answers cite one of our pages without naming Agentcy, up from 12. The clearest case is our own PR software comparison. Gemini and Google AI Mode cited it in five answers and named Agentcy in none of them. The reason was on the page: its opening summary named seven tools and left ours out. Gemini repeated the summary we wrote.
One answer cited the index page itself, for the first time. It didn't name Agentcy, so there was nothing to set aside.
What we changed
We rewrote the opening summaries of the pages assistants cite most, so each says plainly where Agentcy fits: the PR software comparison, the UK media monitoring comparison, the agencies page, the pricing page and our guide to offering AI visibility as an agency. We added answers to questions nobody is winning yet, such as how agencies track time against retainers and how to report to a client each month. And we slowed our own requests to Google, so a rate limit doesn't cost a third of an assistant's answers again.
The rest of the table
Muck Rack stayed first with 196 answers, ahead of Cision (189) and Meltwater (171). Semrush (122) overtook Prowly (113) for fourth, the first change in the order of the top six since the index began. Meltwater gave back most of last week's rise, falling from 192, and PRCoverage roughly halved from 20.
What we take from it
- Check your own collection before you read the result. A third of one assistant's answers were missing this week, and the first-pass number made the fall look nearly twice as large as it was.
- Assistants repeat your summary, so be in it. A comparison page that leaves you out of its opening lines teaches the assistant to leave you out too.
- Gains don't stay won. Gemini named Agentcy for the first time last week and not at all this week. Models change underneath you, as Grok did. Judge a question over a month.
- Named and accurate are two different tests. One new recommendation described our free plan wrongly. A weekly check has to read the answer, not just count it.
Next week
The next run is on 17 November, and the index will show the result whatever it is.
To see what assistants say about your own brand, try the free AI visibility checker, or use AI visibility tracking to run your own buyer questions every week.