Key takeaways
- AI answer sentiment and press sentiment are measured on different text, using different methods. A brand can score well in one and poorly in the other without any change in public opinion.
- AI answers pull from a different source mix than a media monitoring feed. News is only 5.7% of ChatGPT citations, while product pages and landing pages make up about 39%.
- A sentiment number is only comparable if you know the denominator, the label definitions, the source mix, and the time window.
- Measure each AI engine separately, score each mention on its own, and treat answers that omit your brand as absences, not neutral results.
- When the two scores disagree, the gap is a diagnostic. It tells you which sources the AI is reading that your press monitoring never ingested.
The short version of the problem
A brand team opens its media monitoring dashboard and sees net sentiment of +38. Coverage is steady, the last funding round got friendly write-ups, nothing is on fire. Then someone asks ChatGPT "is [brand] good for small teams?" and the answer says the product is "powerful but expensive, with a steep learning curve."
Both readings are probably correct. They are just measuring different things.
Press sentiment asks how journalists and publications write about you. AI answer sentiment asks how a model characterizes you when a buyer asks a question. The second one is synthesized from many sources, only some of which are press, and it gets delivered as one confident paragraph. This guide explains where the two come apart, how to measure each one honestly, and what to do when they disagree.
What each score actually measures
Press sentiment
Traditional brand monitoring collects mentions from news outlets, blogs, social platforms and review sites, then classifies each mention as positive, neutral or negative. Tools like Meltwater, Brandwatch, Talkwalker and Brand24 do this at scale. The unit being scored is a human-authored piece of text that exists independently of any question.

AI answer sentiment
AI answer sentiment is the tone and framing a model applies to your brand inside a generated response to a prompt. It captures whether the model presents your product as solving a problem or as a limitation, and whether it recommends confidently or hedges. It is different from mention rate, which is a visibility question. A brand can show up often and still be framed in language that nudges buyers elsewhere.
The cognizo.ai guide on tracking sentiment in AI answers makes a point worth repeating: because AI responses blend many sources into one narrative, users rarely cross-check the characterization. The framing in a single answer carries more weight than any one review would.

Why the two scores diverge
The most useful way to think about it comes from the BrandMentions sentiment guide: the AI measurement path runs parallel to your monitoring path, not downstream of it. Your press monitor and the AI answer engine both look at the world, but they look at different parts of it and process what they see differently.
A brand can hold a pristine net sentiment score while generated answers describe it with qualifiers drawn from sources the monitoring never ingested. Here are the main reasons that happens.
1. The source mix is different
Promptwatch Data's report on which content types ChatGPT cites the most (Sep 28 to Oct 4, 2026) shows what a model actually reads when it builds an answer. This is citation data, not sentiment data, but it explains a lot.
| Content type | Share of ChatGPT citations |
|---|---|
| Product pages | 22.4% |
| Landing pages | 16.8% |
| Listicles | 13.6% |
| How-tos | 8.7% |
| Documentation | 7.5% |
| General articles | 6.1% |
| Comparisons | 5.9% |
| News | 5.7% |
| FAQ | 3.3% |
| Reviews | 1.7% |
News is a small slice. Product and landing pages, which the brand itself writes, are about 39% together. Editorial content (listicles, how-tos, articles, comparisons, news) adds up to about 43%. So the thing your press monitor watches most closely is a minor direct input to ChatGPT's answer. The thing it watches least, your own site plus roundup articles, drives most of it.
2. Press coverage decays fast in AI answers
Promptwatch's content half-life report found that news pages lose half their citations within about five weeks, while landing pages and docs keep most of theirs for months. A glowing launch story lifts your press score immediately and may fade from AI answers within weeks. A stale pricing page or an old comparison can keep shaping answers long after the news cycle has moved on.
3. Branded prompts pull from a different place
For branded prompts like "is HubSpot good for startups," the mix shifts. Product pages (23.3%) and landing pages (22.8%) still lead, but reviews jump to 5.6%, more than three times their share across all prompts, and social posts reach 3.0%, almost four times. FAQ pages are 4.8% and press releases 2.1%. When a buyer asks directly about you, review sites and community posts matter much more than they do for generic category questions, and those are exactly the sources a news-focused monitor tends to underweight.
4. Different engines read different places
Sentiment in AI answers is not one number, because the engines do not share a source diet. Promptwatch's Social Media Citations in AI Search, November 2026 report (Oct 11 to Nov 9, 2026) shows how lopsided it is:
| Engine | Social share of citations |
|---|---|
| Google AI Overviews | 13.16% |
| Perplexity | 1.94% |
| ChatGPT | 0.64% |
AI Overviews cites social platforms about 20 times as often as ChatGPT. YouTube alone is 6.71% of AI Overviews citations, more than the other six social platforms combined. So if Reddit threads and YouTube reviews are unkind to you, that tone will show up in AI Overviews far more than in ChatGPT, where press and owned pages carry more weight. One caveat: the October and November reports cover different windows (7 days versus 30), so month-over-month changes are not strictly like for like.
5. The source mix can change overnight
On August 14, 2026, reddit.com's share of ChatGPT citations fell from roughly 4% to 0.5%, per Promptwatch's Reddit citation data. Nobody's opinion of any brand changed that day. But the evidence ChatGPT was reading did, and any brand whose reputation lived largely on Reddit would have seen its AI framing move. This is the cleanest example of a score shifting without sentiment shifting.
6. Classification quirks
Sentiment classifiers misread sarcasm, mixed opinions and industry jargon. Adobe's Brand Visibility training material gives the example of a word like "loss," which can be positive in the right context. It also notes the same prompt can return a different tone across users and days. These errors hit both measurement paths, but they hit them differently, because the text is different. Press articles are long and structured. AI answers are short, hedged and full of qualified praise.

Why the number itself might be the problem
Before blaming the sources, check whether you are comparing like with like. BrandMentions makes a point I wish more dashboards followed: a sentiment score is only meaningful when its methodology is visible. The dataset, label definitions, denominator, weighting, source mix and time window all shape the result.
Two specifics matter here.
- Neutral handling. Including or excluding neutral mentions changes the reported result even when the underlying mentions are identical. If your press score drops neutrals and your AI score keeps them, they are not comparable.
- Absences are not neutral. In their worked scorecard, 120 answers that omit the brand are absences. Folding them into a neutral bucket inflates the apparent calm of that surface. If ChatGPT only mentions you in 30% of answers, the sentiment of those 30% is a different question from the 70% where you are invisible.
Also watch the aspect. A mathematically correct score still misleads if the labels concern the wrong brand or the wrong aspect. "Great company culture" in a press piece and "expensive for small teams" in an AI answer are both accurate, and neither cancels the other.
A practical framework for tracking both
Step 1: Keep the two measurement paths separate
Do not blend press sentiment and AI sentiment into a single brand health number. Report them side by side with their own methodology notes. Your press monitor stays a press monitor.
Step 2: Build a fixed prompt set for AI answers
Use a stable set of prompts that mirror how buyers actually ask: category questions, comparison questions, and branded questions. Cognizo's guidance is sound here: a stable prompt set, per-prompt scoring and stored full responses are what let you trace what drove a shift. Weight high-intent prompts more, since a single strongly negative framing in a comparison prompt can outweigh several neutral mentions in low-intent questions.
Step 3: Score each mention, label mixed answers
Classify every mention as positive, neutral or negative, and give mixed responses their own label. "Great product, but pricey" should not be rounded to positive. Attach a short contextual tag (pricing, support, features, reliability) so the score explains itself. A number without a reason is hard to act on.
Step 4: Measure each engine on its own
Track ChatGPT, Google AI Overviews, Perplexity, Gemini and Claude separately. Averaging them hides the pattern from the table above, where one engine leans on social and another leans on owned pages.
Step 5: Keep a change log
A model update, a source-access change or a rewritten query set all move the metric. Log them. When your score drops three points, the first question should be "what changed in the measurement?" before "what changed in the world?"
Step 6: Trace the sentiment back to sources
This is the step most teams skip, and it is where the divergence gets resolved. For every negative or hedged framing, find which URLs the answer cited. That tells you whether the problem sits on your own site, in a roundup article, on a review platform or in a Reddit thread.
Tools for each side of the measurement
The press side is mature. The AI side is newer and the tools differ mostly in how much they help once you find a problem.
| Job | Example tools | What to check |
|---|---|---|
| Press and social sentiment | Meltwater, Brandwatch, Talkwalker, Brand24, Awario | Which sources are ingested, how neutrals are handled |
| AI answer sentiment tracking | Promptwatch, Profound, Peec AI, Otterly.AI, Cognizo | Real interface capture, per-mention scoring, citation tracing |
| Review and reputation sources | Trustpilot, Birdeye, Reputation | Whether they feed the pages AI actually cites |
For AI sentiment, Promptwatch scores sentiment per response by prompt, topic and competitor, and its Brand Narrative feature pulls out the individual claims AI makes about you (such as "affordable for small teams" or "its mobile app lacks features"), grouped under themes like Pricing, Features and Support. Fact Check then lets you mark a claim inaccurate and finds the pages on your site that state it, so you know where the model learned it. That claim-level view is what makes the press-versus-AI gap diagnosable instead of just visible.

Other options in the same category include Profound, Peec AI, Otterly.AI and Cognizo, which tracks visibility, sentiment and owned and earned citations. Most of these stop at monitoring. If you are comparing the category, the GEO software directory at bestgeosoftware.com lists more of them.
Profound

Cognizo

A separate note on Listen Labs: its comparison article groups sentiment tooling into social listening, AI-search monitoring and research-grade diagnosis, and argues that dashboards flag a change without explaining it. That framing is useful even if you never buy primary research. Both a press monitor and an AI tracker tell you what moved. Neither tells you why without a layer of evidence on top.

Reading the gap: four common patterns
Once you have both scores, the combination tells you where to look.
| Press sentiment | AI sentiment | Likely cause | First move |
|---|---|---|---|
| High | Low | AI reads sources your monitor ignores: reviews, forums, old comparison pages, your own outdated pages | Trace citations on the negative answers |
| Low | High | Model leans on owned pages and listicles; press crisis is recent and not yet absorbed | Watch for decay of owned-page advantage; address the press issue |
| High | High but you rarely appear | Sentiment is fine, coverage is the problem | Treat as a visibility gap, not a reputation issue |
| Low | Low | Real reputation problem across sources | Fix the underlying issue, then refresh pages |
The third row is easy to miss. High sentiment on 20% of answers is not the same as healthy AI perception.
What to do when AI sentiment is worse than press
- Find the cited sources behind the negative framing. If it is your own pricing or comparison page, that is the fastest fix.
- Check factual accuracy first. If the claim is wrong, correct the pages that state it. If the claim is accurate but unflattering, that is a product or positioning matter, and no content tweak makes it go away.
- Update owned pages that persist. Promptwatch's half-life data says landing pages and docs stay cited for months, so they are the highest-leverage surface. FAQ and review content matter too for branded prompts.
- Get into the roundups AI already cites. Listicles are 13.6% of ChatGPT citations. Being included in the right ones changes the evidence base.
- Do not chase press alone. News is 5.7% of ChatGPT citations and decays within about five weeks. Press is still worth earning, but it is a weak lever for persistent AI framing.
- Re-measure with the same prompt set. Otherwise you cannot tell whether the fix worked or the model just changed.
If you need help running this as a program rather than a one-off audit, an agency that handles both digital PR and AI search work can cover both halves. 1001 SEO Media does that, and uses Promptwatch for the AI visibility side.
Frequently asked questions
Does negative AI sentiment matter if my brand still appears often?
Yes. Because AI answers synthesize rather than list, buyers rarely see the sources behind an unfavorable characterization, so the framing carries implicit authority. Frequent mentions with hedged language can send buyers to a competitor.
Can I use my press sentiment score as a proxy for AI sentiment?
No. The two paths are parallel. Press is a minor direct input to ChatGPT, and each AI engine weights sources differently. You can use press sentiment as one hypothesis for why an AI score moved, but you cannot substitute one for the other.
Why did my AI sentiment change when nothing happened?
Check your change log. Model updates, shifts in which sources get cited (like the August 14, 2026 Reddit drop in ChatGPT), revised prompts and different time windows all move the score without any change in opinion.
Should neutral mentions count in the score?
It depends, but you must state it. Including or excluding neutrals changes the reported number on identical data. Whatever you choose, apply the same rule to every surface you compare.
Is Promptwatch data a measure of sentiment?
The citation and crawler reports cited above measure which sources feed AI answers, not tone. They explain why sentiment differs by showing what the models read. For sentiment itself, you need per-response scoring on your own prompts.
The bottom line
A press score tells you how publications write about you. An AI answer score tells you how a model summarizes you to a buyer, using a different and faster-changing source mix. They will disagree, and the disagreement is useful once you stop treating it as an error. Track them separately with visible methodology, trace every negative framing back to its cited pages, and fix the sources that actually feed the answer. For a deeper look at how citations work, start with Promptwatch's citations vs mentions guide and its Promptwatch Data reports.

