- Home
- PR Marketing
- Rethinking PR measurement: from press clippings to AI visibility
PR measurement · Article
Rethinking PR measurement: from press clippings to AI visibility.
A clippings report shows where an article appeared and how many people could have seen it. It does not show whether that article shapes the answers of ChatGPT, Gemini or Perplexity. That matters, because many buyers now let an AI explain a market to them. Here is how we, as an agency, measure press work anew: with a question set, a monthly log and a report that connects coverage to AI answers.
- Reach, share of voice and placement counts still matter, yet none of them reveals whether coverage influences AI answers.
- The second layer records mentions, cited sources, accuracy of the description and competitors named.
- The basis is a fixed set of 25 to 50 buyer questions, put to the major AI systems monthly.
- AI answers fluctuate. Only the development across months counts, never a single reading.
- No one can promise a mention. What you can measure is whether your brand’s position changes.
What the clippings report no longer shows.
Buyers increasingly start their research in a chat window. They ask ChatGPT, Gemini, Perplexity, Microsoft Copilot or Google’s AI features, from AI Overviews to AI Mode, to explain a market, set providers side by side and build a shortlist. Many never click through to the trade outlet itself.
Almost halved: when an AI summary sits above Google’s results, only 8 percent of visits end with a click on a conventional search result; without one, the figure is 15 percent. That comes from an analysis of real browsing data that the Pew Research Center published in 2025.
A title’s circulation therefore says less and less about its human readership, while AI systems work through each piece carefully and keep returning to it. Independent editorial coverage, known as earned media, becomes more valuable as a result, since answer engines lean strongly on sources that do not come from the company itself. So the answer is not less PR, but better aimed PR, measured differently. Reach, impressions, share of voice, placement numbers and key-message pickup stay useful, yet none of them tells you what an AI says about your company.
The second layer: AI visibility.
So we do not replace the classic metrics, we add a second layer that measures what happens in the answers themselves:
| What is recorded | The question behind it | |
|---|---|---|
| Mention | Whether the brand turns up in the answer, and if so, where | Are we on the shortlist? |
| Sources | Which outlets and pages the AI cites, your own coverage included | Is our press work reaching the answer? |
| Accuracy | How correctly the AI describes company, category and offer | Are we understood correctly? |
| Competitors | Which other providers are named alongside | Who are we compared with? |
Only together do the four give a full picture: a brand can be named often and still be described wrongly, or appear without any of its own coverage being cited.
What kind of question is asked makes a difference as well: news and market developments are answered mostly from news sites and the trade press, “how do I” questions more often from reference works and providers’ own websites, guides and documentation. So the measurement also shows whether coverage or your own site has to do the work.
Step 1: The question set.
It starts with a list, not software: 25 to 50 buyer questions, worded the way customers actually phrase them. We write them together with the company, drawing on sales, customer service, enquiries and trade fair conversations. The questions fall into three stages:
- AwarenessHow someone gets to know a market: which methods exist and what they are suited to.
- ComparisonProviders, technologies or approaches are weighed against each other.
- SelectionShortly before the decision: which provider fits a specific requirement.
Each question also carries its type, industry topic or “how do I”, so a gap can later be traced to coverage or to your own pages. The set then stays fixed, since only identical wording gives comparable results; new questions go into a separate block. And every company writes its own: a copied list measures someone else’s market.
Step 2: The monthly log.
Once a month, we put every question to the AI systems that matter to the target group, usually ChatGPT, Gemini, Perplexity, Copilot and the AI features in Google Search. For each answer, the log records five points:
- Named: yes or no
- Position: where the brand appears in the answer
- Sources: which outlets and pages are cited
- Accuracy: whether name, category and offer are correct, and if not, what is wrong
- Competitors: who else is named
The date, the system and the exact wording of the question are noted too. A simple table will do at first; dedicated tracking tools can take on larger volumes later.
One quirk matters: ask the same question twice and an AI system may well answer differently. Any single reading is therefore a snapshot. We look at how the figures move across several months and only draw conclusions once a pattern repeats.
Step 3: The report.
The report condenses the log and sets it next to the month’s press work, always in the same cycle:
- Ask the questions
- Log the answers
- Match against coverage
- Adjust the media list
- Align the description
It sets mentions, cited sources and errors in the description against the month’s placements, names competitors with their sources and plans the next month.
Re-sorting the media list
The key outcome is a new list of target outlets, ranked by how often an outlet is cited for the topic. Reach and domain authority are poor guides here: for precise questions, AI systems often cite small specialist titles, whereas broad outlets with general reporting seldom show up. Coverage counts most when it answers a real buyer question, defines the brand clearly, is recent and is not a one-off. For current questions, AI systems favour fresh content, while substantial older pieces still serve as the base.
One description everywhere
Defining the brand clearly means saying who the company is, its category, whom it serves, and one plain, checkable fact. A quote full of excitement about a new partnership contains no fact an answer engine could pass on. The same description belongs in press texts, on your own site, in team profiles and in directory entries. If these places disagree, AI systems lose confidence and may fall back on how a competitor describes the market, or on stale claims.
An example: the pump maker from the Sauerland.
A mid-sized maker of industrial pumps in Germany’s Sauerland region, let us call it Brakenfeld Pumpentechnik, supplies dairies and breweries. Its clippings report looks good, including a profile in a major business magazine.
Question set: With sales and customer service, the team writes 40 questions, such as which pump design suits viscous foods or who offers pumps that can be cleaned while the line keeps running.
First log: In selection questions, Brakenfeld is hardly named. The answers cite two small food technology journals and an industry association guide, none of which the company had ever contacted. The magazine profile appears in no answer. One AI also mentions a product line discontinued years ago, taken from an old directory entry.
Actions: The two journals move to the top of the media list. The first trade article answers a question from the set and states name, category, audience and one plain proof point: the pumps can be cleaned without being taken apart. Directory entries, team profiles and the “About us” page get the same descriptive sentence.
Report: From the second month, each placement is shown alongside whether it is cited and whether mentions and position change, judged over several months.
What the example does not promise: a place in every AI answer after one article. What it delivers: the team knows where its buyers get their information and measures every placement against that.
What to ask of your agency.
A short self-check shows whether your press work already works for AI answers. Tick what is true today:
Where most boxes remain unticked, your press work is still missing the places where buyers make up their minds today. In that case, three requests to your agency make sense: pitches backed by data on which outlets are cited for your subject, story ideas rooted in buyers’ AI questions, and a monthly account of whether coverage shifts where your brand stands in AI answers.
At NexTao, this review is part of GEO care: every month, we check in selected AI search services how your company appears for typical customer questions and report on it clearly. GEO Starter costs €290, GEO Business with a video for every advice article €490 and GEO Premium with social media management on top €1,490 a month, plus a one-off €490 setup, waived with the annual package at €2,900, €4,900 or €14,900. All prices excl. VAT.
For regular articles in magazines that fit your topic there is pr4geo, our brand for PR aimed at AI answers; for “how do I” questions, website4geo builds websites that people and AI systems read well. No agency can promise you a place in AI answers, and that includes us. We promise the work and honest measurement.
Frequently asked questions.
How many questions does the set need?
To begin with, 25 to 50, spread over all three stages: awareness, comparison, selection. Keeping the wording unchanged from month to month matters more than the exact number.
Why is a single query not enough?
Because the same question can produce a different answer on the next attempt. Only a pattern that holds across several months is reliable.
Can an agency guarantee that ChatGPT will name us?
No. The AI system decides which sources it draws on. What can be promised is the work, regular measurement and an honest report.
