Pressfront
Research · October 2026
A press release helps AI know you by name. It did nothing for "best app" questions.
We sent out a real press release, then asked ChatGPT, Gemini and Claude the questions a customer would actually type.
0 of 48"best app for..." answers that used the press release
14 of 23answers about the company by name where AI read a copy of the release
-
We sent out one real press release
- For TableCash, a group payments app from the founder behind Pressfront. Our own product, so nobody else's data.
- A $149 newswire package, published September 6, 2026.
- The newswire's report listed 537 sites carrying it.
-
Asked about the company by name, AI read the release
- Questions like "What is TableCash?" and "Is TableCash legit?"
- A copy of the release was a source in 4 of 8 ChatGPT answers and 3 of 8 Gemini answers.
- It turned up in Claude's search results 7 of 7 times.
- The company's own site and app store pages were used almost every time anyway. The release backed them up. It didn't carry them.
-
Asked the everyday questions, the release never came up
- 8 questions a real customer asks, none naming the company: "best app to pool money for a group trip", "Splitwise alternatives that actually collect the money", and so on.
- 3 assistants, 2 runs each: 48 answers.
- The release was used 0 times. The company was named 2 times, both by ChatGPT, both from the company's own page about group trips.
-
What AI used instead: the competitors' own pages
- One page per use case. Cheddar Up has separate pages for group gifts, group trips and team payments. Each one got quoted for the matching question.
- Honest "best app for X" guides on their own blogs. SquadTrip's guides were quoted in 9 of 16 Claude answers.
- Lists on other sites that compare apps in the category.
Not one
of the copies AI actually read (on sites like aijourn.com and news.marketersmedia.com) was in the newswire's 537-link report. If you check a release by its report, you're checking the wrong sites.
What to do with this
- A press release is good for one job: when someone asks AI about you by name, it has something to read.
- Don't buy one to show up in "best X" answers. In our test that number was zero.
- For "best X" questions, write the page yourself. One page per thing people use you for, plus an honest comparison that names your competitors.
- Then get onto other sites' lists for your category. Published research finds that is the biggest lever for small brands.
- Measure the right thing: count answers that name you, not answers that link to some page of yours.
The longer version, with the outside research behind each step: how to show up in AI answers.
The numbers, per assistant
| Assistant | Named the company | Used the release |
|---|
| ChatGPT (gpt-4.1-mini, web search on) | 2 of 16 | 0 of 16 |
| Gemini (3.5 Flash, Google Search on) | 0 of 16 | 0 of 16 |
| Claude (Opus 5, web search on) | 0 of 16 | 0 of 16 |
Everyday "best app" questions only. 2 runs per question per assistant.
How we did it
- Every question was asked through each assistant's own API with web search turned on and a US location, 2 times each. Answers change from run to run, so one screenshot proves nothing.
- "Used the release" means a copy of the release was a cited source. We matched on the text of the release, not on the newswire's list of sites, because the list missed the copies AI used.
- "Named" means the company's name appears in the answer text. A link alone doesn't count.
- Claude got a short instruction to search first and not ask questions back, to match how the app behaves.
- Dates: release published September 6, 2026. Name questions and everyday questions asked in October 2026.
What this test can't tell you
- It's one release, for one app, in one category. A different category could behave differently.
- One of the 8 name-lookup answers from Claude failed, which is why its count is out of 7.
- AI answers change week to week. These are October 2026 numbers.
- The test was run by the founder's own team, on the founder's own app. We publish the method so anyone can repeat it.
Journalists and researchers: the figures on this page are free to cite with a link to pressfront.co.