The four questions to ask before buying any AI PR tool
Every AI tool sold into public relations demos well. That is not a criticism of the vendors so much as a description of the category: the tasks are text-shaped, the outputs are short, and a language model is very good at producing something that looks like a pitch, a release or a summary. The demo is the easy part.
The difficulty starts about three weeks in, when the tool meets a client whose name is also a common noun, a journalist who moved outlets in March, and a chief executive who wants to know why the sentiment score moved. Almost every failure we have seen falls into one of four buckets, and all four can be tested before you sign anything.
Where did this data come from, and when
Any tool that touches journalist contacts, coverage or outlet information is making a claim about the world, and that claim decays. Newsrooms churn constantly. A contact record that was correct in January may be wrong by April, and nothing about the interface will tell you which.
So ask for the verification date on individual records, not the size of the database. A vendor who can show you when a specific contact was last checked has built something maintainable. A vendor who answers with a total contact count has told you they do not track it.
The same question applies to coverage matching. If a tool says a journalist covers your sector, ask what it read to decide that. The good answers cite articles. The weak ones cite a beat tag somebody assigned years ago.
What happens when it does not know
This is the single most useful question to ask in a demo, and it is rarely on the script. Push the tool towards the edge of its competence — a small regional outlet, a niche trade title, a company with an ambiguous name — and watch what it does.
A well-built tool degrades honestly. It returns fewer results, flags low confidence, or says it has nothing. A poorly built one produces the same fluent, confident output it produces for everything else, because the underlying model will always produce something.
That difference matters more in PR than in most fields. A fabricated statistic in a press release is not an inconvenience, it is a correction, and possibly a regulatory problem. The tools worth paying for are the ones that would rather return less.
Can you reconstruct the number
Measurement products almost all ship a proprietary score. An index, a health rating, a visibility number. Very few publish how it is calculated, and the reason is usually that the calculation is not very interesting.
The test is whether you could explain a movement in that number to a client who was not in the room. If the score fell eleven points this month and nobody at the vendor can tell you which inputs moved, you have not bought measurement. You have bought a chart.
Ask for the methodology in writing. A vendor confident in their approach will send it. This is also the fastest way to find out whether advertising value equivalency is still lurking somewhere in the stack, which it often is, because clients recognise a currency figure even though the industry abandoned the metric years ago.
What does it cost when it works
The pricing question that catches people out is not the licence fee. It is what happens on success.
Distribution products often price per release, monitoring per mention or per keyword, outreach per seat and per send. Each of those is a charge that scales with you doing more of the thing you bought the tool to do. That is a reasonable commercial model and a bad surprise if you have budgeted on the headline number.
Model the second year, not the first. Take the volume you would run if the tool worked exactly as promised, and price that. If the answer is uncomfortable, the tool is not priced for your use case, however good the demo was.
What this looks like in practice
None of these four questions require technical knowledge, and all four are answerable in a sales call. Most of them feed into how we score.
Where the data comes from, and whether you can check and correct what the tool gives you, is reliability. How a tool behaves when it hits the edge of what it knows tells you how good the AI behind it really is.
Price is not one of the three checks. It still decides whether a tool is worth buying.
The tools that come out well here are rarely the ones with the most features. They are the ones that tell you what they do not know.