For PR and comms teams
What AI answers change for PR, and what they do not
People and journalists now ask AI assistants about organisations. What that changes for a PR team, what it does not, and three things you can do yourself.
One AI answer is one draw, not a measurement. This website sells nothing.
Marketing and sales teams can use the toolkit too: its market questions show what an AI assistant tells a buyer who asks about your field.
What has changed
Many people use these tools. In 2025, 32% of people aged 16 to 74 in Germany, 39% in Austria, 30% in Hungary and 47% in Switzerland used generative AI tools in the previous three months. The EU-27 figure is 33% (Eurostat). This is general use, not research about companies.
Journalists use them too. 70% of journalists surveyed in Germany, Austria and Switzerland use AI tools in their everyday editorial work (news aktuell, 2025; 775 answered this question). news aktuell distributes press releases, so this is a vendor survey.
Journalism is cited far more than press releases, in one vendor's data. In that vendor's log data, about 27% of the links AI assistants cited were journalism and 1.1% were press releases (Muck Rack, May 2026).
What it does not mean
- Not a takeover. Across the markets surveyed, 10% use AI chatbots for news each week. In Germany it is 5% (Reuters Institute, 2026).
- Not the end of the press release. 1.1% is one vendor's share of cited links, not a measure of influence. Nothing here shows the press release has stopped mattering; facts that are correct, dated and easy to check are the ones an answer can be checked against. Write for people, not wording for machines.
- Not a new score to buy. A score alone does not settle what an AI answer says, and nothing on these pages needs buying.
- Not a rival to your measurement framework. A reading of AI answers records facts and sources. It is not reach, sentiment or return, and it sits beside the outcome measures you already use.
- Not a crisis every time. One wrong answer that one person saw once is not yet an issue.
What it means for a PR team
Your organisation has a second description that nobody approved. An AI answer describes you, your releases and your spokespeople in its own words, from sources it picks. It can be wrong, and it can change from one ask to the next.
The question gets wider. It used to be: was it published, and where? Now there is a second one: what does an answer say about us afterwards, and is it right?
It is about the whole organisation, not only its press. Buyers, journalists, job candidates, critics and officials may all ask. Ask the way a critic would, too.
It is listening, not winning. The aim is that people who ask are told the truth, checked against your own record.
One sentence for your board
"AI answers are a new place where we are described. Before we spend anything, we check what they say: more than once, in writing, against our own source."
Five things teams ask first
- Who owns this? Comms, HR, legal, marketing and IT each tend to think another team does. Name one owner, and say which decision the reading is for.
- When is a wrong answer an issue? A test for your own judgement, not a number. It comes back when you ask again. It is harmful, or about a named person. You can trace it to a source you can fix. Then fix your own sources first.
- What should we not paste in? Ask about public facts only. Do not paste unreleased or embargoed releases, client material or personal data into an assistant. Keep it safe.
- What does this not ask of us? No gaming, no text written to steer engines, no real personal data in tests.
- What about the legal side? Ask your own counsel, for example how long to keep a log of answers and who may see it. Makefield studies what AI answers say. It does not give legal advice.
What nobody here can tell you yet
- How often people in Germany, Austria, Switzerland or Hungary ask AI assistants about companies. We found no checked figure.
- How often AI answers about organisations are wrong. We found no checked figure. This page reports none of Makefield's own results.
- How Hungarian journalists use AI. The journalist survey above covers Germany, Austria and Switzerland only.
- Whether an AI answer is now a journalist's first background check, before they call you. It is an open question.
Evidence and its limits
Each line says who asked, how many, when, and who paid.
- Eurostat, dataset isoc_ai_iaiu: official statistics, survey year 2025, people aged 16 to 74. Limit: general use, not company research.
- Reuters Institute Digital News Report 2026: YouGov online survey, about 2,000 people per market, fieldwork early 2026, published June 2026. Independent. Limit: news use, not brand research; no figure for Austria, Switzerland or Hungary in the text.
- news aktuell Medien-Trendmonitor 2025: online survey of journalists in Germany, Austria and Switzerland, May to June 2025, 775 answered the AI question. Label: vendor survey; not a random sample.
- Muck Rack, "What Is AI Reading?", May 2026: more than 25 million links in answers from three AI assistants. Label: vendor log data; Muck Rack sells PR software. Prompt count and dates are not stated.
- IAB, "Measuring Visibility in the AI Era", August 2026, page 32: "Single-response measurement is not measurement."
- Makefield's floor test: answers to the same question from different rounds were less alike than repeats within one round in 85 of 103 pairs. It compared wording, not facts.
Go deeper
- One AI question, two answers: a five-minute test.
- Audit your AI visibility in an afternoon
- How to read your own AI-visibility audit
- AI got a fact about you wrong
- Who AI cites
- Why Makefield exists
How to cite
Denes Csaszar, "What AI answers change for PR, and what they do not", Makefield, makefield.io/pr, last checked 26 October 2026, version 1.
Last checked 26 October 2026. What changed: version 1, first published 28 October 2026.
Makefield is an independent research project on AI answer accuracy, based in Berlin, Germany. This website sells nothing.