Get a source address for every claim AI makes about your brand. Ten questions checked by hand cost EUR 450 to EUR 900.
Your job has always been what gets repeated when nobody from the company is in the room. That used to mean journalists and analysts. Now a model answers first. It has a view on your pricing, your service and whatever went wrong two years ago, it says all of it with total confidence, and some of it is wrong. You find out which parts when a journalist reads one back to you on a call, which is the most expensive moment there is.
Then there is the half of the job that has never had an after-shot. The campaign lands, the coverage arrives, someone senior asks what changed. You show the clippings. Nobody has ever shown the answer itself moving.
What it costs today
Comms and analyst time at EUR 90 an hour.
| Job, by hand | Arithmetic | Cost |
|---|---|---|
| Check one risky question: ask it three times, count the mentions, tag sentiment | 30 to 60 min at EUR 90/h | EUR 45 to EUR 90 |
| Ten questions, three rolls each | 5 to 10 h at EUR 90 | EUR 450 to EUR 900 |
| Prove the answer moves: one question, ten rolls | 1.5 to 3 h at EUR 90 | EUR 135 to EUR 270 |
| One round on a person: 18 probes in two languages, about 200 citations classified and graded | 6 to 10 h at EUR 90 | EUR 540 to EUR 900 |
| Source and vet paid placements, one campaign | 5 to 15 h at EUR 90 | EUR 450 to EUR 1,350 |
| Rebuild the tracker that proves the number moved | 3 to 6 h a month at EUR 90 | EUR 270 to EUR 540 a month, EUR 3,240 to EUR 6,480 a year |
Bought in, it costs more. An executive reputation audit from a comms agency runs EUR 2,000 to EUR 6,000 one-off. A link or PR agency charges EUR 500 to EUR 2,000 a month to do the editor mailing. A personal-brand retainer of EUR 1,500 to EUR 4,000 a month is allocated on instinct today.
Four of these five playbooks run on Marketer at EUR 196 a month. At EUR 90 a question, the third question you audit clears that plan. Dice Roller sits on Strategist at EUR 496, covered by the second question a month at EUR 270 and by the fourth at EUR 135.
The workflow
Five videos, 27 minutes, in the order a comms team would run them.
1. Catch what AI invents about your company · 5:36 · Marketer
Start with the question you would hate to see in print. Add it as a custom prompt and run it three times in Search mode. The claims repeated on all three rolls are your public position now, whoever wrote them; the claims that surfaced once and vanished are the soft ones, where the wrong ones live. Brand Sentiment gives you the exact sentences behind the negative group. Sources gives every cited page and which rolls used it, so a wrong claim gets an address. Export as PDF, Markdown or CSV and the file goes to legal.
2. Show how unstable the AI answer really is · 4:44 · Strategist
The budget ask dies when a colleague opens ChatGPT on a phone, types the category question once, sees the brand and decides the problem is imaginary. Ten rolls of that question give ten different lists. Every brand named is sorted into four bands by how many rolls carried it: Always at 80% and above, Frequent from 50, Occasional from 20, Rare below. A rival in Always beside your own name in Occasional is the slide. Message Stability shows the sentences the model repeats every time, which is the copy you have to own.
3. What AI says about a person, and where the budget goes · 6:10 · Marketer
Your CEO gets shortlisted by a model before anyone opens LinkedIn. A person is a project like any other here: a name and a live domain. You get the score and its six parts, then the places a model went to learn about them, grouped by source type, with sentiment on each and a filter by theme. Dmitrij ran it on himself first and came back at 58, with a research profile, an academic paper and a conference speaker page behind the answer. Not LinkedIn, where he had posted for years. That is the evidence for moving a retainer, and every citation exports as a CSV row.
4. Buy only the placements that feed AI answers · 5:20 · Marketer
Models quote sources, so the placement question is a PR question. Set a budget, pick an article the project already produced, and one click returns a ranked shortlist with domain rating, monthly traffic, country and price on every card, plus the paragraph explaining the picks. The cart drawer is already your media plan, and confirming sends a real brief to a real editorial desk.
5. Prove the needle moved, with a chart · 5:33 · Marketer
The after-shot. One line with a date range on it, the seven parts the score is built from with the weight each contributed, sentiment broken out theme by theme with a signed score, and an export listing every cited page grouped positive, neutral and negative. One image for the board slide, one file for the person who argues with the method.
What this does not do
Nothing watches the answer for you. There is no monitoring, no alerting, no schedule and no email when something changes. Every run is a manual click. Scans are capped at one per model per calendar month, so a trend chart fills up one point a month.
Gemini, ChatGPT and Grok. Those three, and one model per run. Claude and Perplexity are outside every number on these screens.
The placement shortlist does not know which portals AI already quotes. It ranks on domain rating, monthly traffic, topical closeness and budget fit. No model output sits behind that ranking, and any tool claiming otherwise owes you its evidence. You still pay the publisher for the slot.
We measure what the models say. We do not promise a position. There is no correction button and no way to make a model retract. A wrong claim gets a source address; fixing that page is still your work. The scores here are weighted composites between 0 and 1 shown as percentages, so read the raw answers underneath before you act on one.
Start here
Watch Catch what AI invents about your company. It is 5:36, it needs one finished scan, and it ends in a file you can forward without explaining the method first.
If the wrong claim comes back on all three rolls, you have found a public position nobody on your team wrote, and you found it before a journalist did. Then watch Prove the needle moved, with a chart, because the fix you make next is the one you will be asked to evidence.