Once a release is live, the obvious question is whether it made any difference to what AI engines say. The tempting answer is to open ChatGPT, type the brand name, and screenshot whatever comes back. We understand the instinct, but the result is close to meaningless: the same question can produce a different answer an hour later, and a brand-name query is the easiest question an engine will ever get.
At Indian PR Distribution (IPRD)® we treat AI answers as something to measure, not to sample. This is how we do it, at the level of principle. The detailed question sets and scoring are built per client and stay private.
Key takeaways
- AI answers vary from run to run, so a single screenshot proves very little either way.
- Fix the questions before the campaign: brand, news and category questions, written the way buyers ask them.
- Run the same set across several engines, more than once, before and after.
- Measure a few plain things: mentioned or not, described correctly or not, which sources were cited, and who else appeared.
- Report gaps as clearly as gains. The gaps decide what the next phase should be.
Fix the questions before the campaign
The first step happens before anything is published. We agree a question set with the client, in three groups:
- Brand questions: what the company is, who runs it, where it operates.
- News questions: what it announced, raised, launched or won.
- Category questions: the best options, alternatives and comparisons a buyer would ask about, without naming the brand.
The questions are written the way people actually ask them, including the Indian context (cities, regulations, languages) where it matters. Fixing them in advance stops anyone, including us, from choosing flattering questions afterwards.
Run the same set, more than once, in several engines
We run the full set across the engines buyers use most: ChatGPT, Gemini, Perplexity, Google AI Overviews and Copilot, and others where the audience calls for it. Each question is asked more than once, because answers vary, and we record what comes back as text, not screenshots.
A baseline run happens before the campaign. Follow-up runs happen after it, at intervals agreed upfront, so we can see change over time rather than a single moment.
Measure a few plain things
It is easy to invent complicated scores. We prefer a handful of measures anyone can understand:
| Measure | What it tells you |
|---|---|
| Mention rate | How often the brand appears across the question set |
| Accuracy | Whether the description, facts and figures are right |
| Sources cited | Which pages the engines lean on, and whether campaign coverage is among them |
| Message adoption | Whether the key idea from the release is repeated |
| Share of voice | Who else appears in category answers, and how often |
What the numbers looked like in practice
Two examples, by descriptor. For a new-age tech platform, after a largely sponsored campaign, the brand was named in 80% of the AI observations we tracked, and 58% of the messages we tracked were picked up by the engines. On open category questions, though, it appeared in about one answer in nine, alongside seven to eleven rivals.
For a sustainability e-commerce platform, the brand appeared in 60% of tracked answers, and a framework named in the release was repeated by four of the five engines tested. Category visibility sat at roughly a quarter of answers, and the answers showed a risk of the brand being confused with an overseas namesake, which became a specific fix in the next phase.
In both cases the gaps were as useful as the gains. They showed which questions sponsored copies had already won and which needed independent coverage. That is the route decision again, made with evidence rather than instinct.
Report the gaps as clearly as the gains
Every follow-up report shows the baseline, the latest run, the change, and what we recommend next. Where an engine is wrong about a brand, we say which sources seem to be driving it. Where a category question has not moved, we say so rather than pick a better one.
We also say what the method cannot do. It cannot prove cause and effect with certainty, because engines change and so does the wider web. It can show, consistently and repeatably, whether the answers buyers get have moved in the direction the campaign was meant to move them.
Start with a baseline
If you have never checked what AI engines say about your brand, a baseline is the sensible first step. Our AI visibility audit runs the method above once, across the main engines, and returns the findings with a recommended route. If you then run a campaign, the audit becomes the "before" that everything is measured against.
Frequently asked questions
Can I just ask ChatGPT about my brand to see if PR worked?
You can, but one answer proves little. AI answers vary from run to run, and a brand-name query is the easiest question an engine gets. A fixed question set, run several times across several engines, gives a far more reliable picture.
Which AI engines do you check?
Usually ChatGPT, Gemini, Perplexity, Google AI Overviews and Copilot, with others added where the audience uses them.
Do you share the questions you use?
Each client sees and agrees their own question set before the campaign. We do not publish question sets or scoring methods, because they are built for each brand and category.
Can you prove a press release caused a change in AI answers?
Not with certainty, because engines and the wider web change too. We can show, repeatably, how answers moved against a baseline and which sources the engines cited, including campaign coverage.
