Updated on 19 November 2026, by João Leite, Founder of murmur.
To find out whether AI speaks well or badly of your brand, ask the buyer’s questions and read how each answer describes the company. Do this in ChatGPT, Claude, Perplexity and Google AIO, the summary that Google’s AI puts at the top of many searches, with questions that include the brand name and questions that do not. Monitoring tools call this tone sentiment. Peec AI, which sells AI monitoring, gives the tone of answers that mention the brand a score from 0 to 100, and every other tool does the math its own way.
- The tone is read in the words of the answer. An answer can describe the company with praise, with a dry fact or with criticism, and that is what the tools call positive, neutral and negative sentiment.
- Each tool counts sentiment in its own way. Semrush shows the percentage of favorable and general mentions; Peec AI gives a score from 0 to 100, where most brands sit between 65 and 85; Profound shows whether AI portrays the brand positively or negatively and which sources shape that story.
- Tone and presence are two separate calculations. A brand can appear in many answers and be described with criticism, and Semrush cross-references the two to find that case.
- The tone comes from what AI finds about the company. In a Peec AI study from August 2026, with one-word English names, the meaning of the name changed the description only where AI had little information about the company.
- The reading is repeated over time. The same question changes its answer, and in ChatGPT the location of the person asking also feeds into the search.
Finding out whether AI speaks well or badly of your brand starts with the buyer’s questions
Finding out whether AI speaks well or badly of your brand starts with the questions the buyer asks, not with a single question about the brand. Questions without the company name show how AI describes it when it recommends it among the others, or whether it leaves it out. Questions with the name show what AI says when someone has already heard of it and wants to check, such as “a empresa X é boa?” (“is company X good?”) or “a empresa X é confiável?” (“is company X trustworthy?”). The two lists stay separate, and the guide on how to build the list of questions to measure the brand shows how to set aside the ones that include the brand name.
Tone is different from brand presence. Presence is how many answers the brand appears in, divided by the total number of answers, and in English this calculation is called share of voice. Tone is the other half: how the brand appears when it appears. What presence shows, and what it fails to show, is in the share of voice guide.
Brand sentiment is the tone of the words with which AI describes the company
Brand sentiment is the tone of the words with which AI describes the company in an answer. In Peec AI’s documentation, sentiment measures the overall tone of AI answers when they mention the brand, and a higher score indicates more positive language and associations. The same page splits tone into three: positive, neutral, which is factual language with little emotional tone, and negative, which is language of criticism, concern or negative association.
In a manual reading, the three categories look like this. An answer that calls the company a good fit for a type of customer, or that highlights its customer service, leans positive. One that lists what it sells, where it is and how much it charges, without judging, stays neutral. One that talks about complaints, delays or high prices, or that puts the brand next to a problem, leans negative.
Semrush, which sells an SEO tool, uses two other words. In its brand performance reports, the sentiment associated with the brand in AI conversations is split into favorable and general, and the number is the percentage of mentions of each type.
Each tool counts sentiment in its own way: Semrush, Peec AI and Profound
Each tool counts sentiment in its own way, and one tool’s score cannot be read on another’s scale. All three describe the calculation on their own page, read on 9 November 2026.
| what is being compared | Semrush | Peec AI | Profound |
|---|---|---|---|
| who it is | sells an SEO tool | sells AI monitoring | sells AI monitoring |
| what the calculation looks at | the brand’s mentions in AI conversations | the overall tone of answers that mention the brand | how AI portrays the brand |
| how the result comes out | percentage of favorable and general mentions | score from 0 to 100 | positive or negative portrayal |
| the tone categories | favorable and general | positive, neutral and negative | positive and negative |
| what appears alongside | the chart that cross-references presence and sentiment | the common range, from 65 to 85 | the narratives gaining ground and the sources that shape the story |
Semrush’s help page on brand performance reports says the presence versus sentiment chart is for “identifying brands with high visibility but poor sentiment”, that is, finding brands with a lot of visibility and bad sentiment. In Peec AI, most scores sit between 65 and 85, on a scale from 0 to 100. Profound shows whether AI portrays the brand positively or negatively, which narratives gain ground and which sources shape the story.
Semrush’s percentage and Peec AI’s score are different calculations and are not comparable. A sentiment score only says something with the scale and the tool next to it. The visibility scores of these tools, which count presence and not tone, are in the guide on what each AI visibility score counts.
How to read whether AI speaks well or badly of your brand in six steps
To read whether AI speaks well or badly of your brand without a tool, follow six steps and keep every answer.
- Separate the buyer’s questions without the brand name from the ones that include the name, into two lists, using the words the buyer uses.
- Ask each question in a new conversation, without the memory of previous conversations.
- Repeat the questions in ChatGPT, Claude, Perplexity and Google AIO, because each AI writes its answer with its own sources.
- Note the words each answer uses to describe the company and the source of each sentence, and mark whether the tone is positive, neutral or negative.
- Repeat the same questions over time, because the same question gets a different answer: in a study by SparkToro from November and December 2025, the chance of two answers from ChatGPT or Google’s AI bringing the same list of brands was below 1%. And in ChatGPT, the location of the person asking is also factored into the search.
- Compare with competitors on the same questions, to know whether the tone belongs to the brand or to the way AI talks about the whole sector.
Location carries weight in ChatGPT in a way that OpenAI describes. OpenAI’s help page on search in ChatGPT says it collects approximate location from the internet address of the person asking and may pass it to partner search engines to improve results. That is why the answer a salesperson sees in Curitiba may not be the one the buyer sees in Recife, and the comparison is valid between readings made from the same place.
Step 6 avoids a common mistake. If AI describes every company in the sector with the same caveat, the caveat belongs to the sector, not to your brand. If only your brand appears with criticism, the tone belongs to it, and it is worth going after the source.
Where the tone comes from: the pages AI read and what it knows about the company
The tone of the answer comes from what AI read about the company and from how much it knows about it. When the answer cites sources, the cited pages are the first place to look for where each sentence about the brand came from, and that is why step 4 notes the source. Profound is one of the tools that show which sources shape the brand’s story.
When the AI knows little, other things take the place of fact. In August 2026, Jan Ehrlinspiel, of Peec AI, tested whether the meaning of a name changes the description the AI gives of a brand, using one-word English names and a list of about 14 thousand English words rated by people, published by Warriner and colleagues in 2013. The conclusion, in the Peec AI study on brand names in AI descriptions, is that “the effect lives only where the model lacks information, and disappears once a brand is known.” In that study, the meaning of the name mattered only where the AI had little information about the company, and it disappeared when the brand was known.
Bad tone, wrong information and bad rating call for different paths
Bad tone, wrong information and bad rating look like the same problem, but each one is solved in a different place. The answer that carries a wrong fact about the company, such as a price, an address or a service it does not sell, calls for fixing the fact, from the record of the answer back to the source, as the guide to correcting wrong information shows. The answer that repeats what customers write on review sites leads to the guide on reviews and Reclame Aqui. For bad tone with no wrong fact and no review behind it, the path is to read the sources of the answer and publish what is missing about the company.
murmur works on the source of the tone: how ChatGPT describes Mindo
The tone of the answer comes from what the AI finds about the company, and that source is what murmur works on. murmur, a GEO agency in Brazil that places brands in the answers of ChatGPT, Claude, Google AIO and other AI agents, writes the pages with the answers the buyer is looking for and helps correct and position the brand the right way in AI models. From the diagnostic onward, a specialist stays with the brand: they know the company and its market, read the results and follow every change closely.
The Mindo case study, a B2B motion design studio in São Paulo, shows the tone that comes out of this work. In the July 2026 answers, ChatGPT describes Mindo as an option for original 2D animated video and corporate presentation. These are the words of its positioning: Mindo serves companies on two fronts, corporate presentation and animated video.
In murmur’s service, the answer to each question is stored in the dashboard, and the words of each AI model can be reread week by week.
Frequently asked questions
Does a high sentiment score mean the AI recommends the brand?
No: the sentiment score is about the tone of the answers that mention the brand, and the recommendation depends on the brand appearing in the answers. They are two different figures, and that is why Semrush combines presence and sentiment in a chart, which shows brands with a lot of visibility and bad sentiment. A brand can have good tone in the few answers where it appears and still appear little. Presence is covered in the guide on share of voice in AI.
What sentiment score do the tools consider common?
In Peec AI, most sentiment scores fall between 65 and 85, on a scale from 0 to 100, and a higher score indicates more positive language and associations. Semrush does not use that scale: it shows the percentage of favorable and general mentions of the brand. The 65 to 85 range applies only to the Peec AI score, and a score can only be read with the scale of the tool that calculated it.
Does the tone change when the AI comes to know more about the company?
In an August 2026 Peec AI study, with one-word English brand names, the meaning of the name changed the brand description only where the AI had little information about the company, and the effect disappeared when the brand was known. What it shows is that, in that test, information about the company took the place of the name in the description.
Do customer reviews change the AI’s tone about the brand?
Reviews enter the tone when the AI uses review sites as the source of the answer, and the sentence about the brand repeats what customers wrote there. What those sites bring to the answer, and the reputation Reclame Aqui calculates, has its own guide.
Where to start
To see today’s tone, start with the answers AI models give to people looking for what your company sells. In murmur’s free diagnostic, all you need is your site’s URL: in less than two minutes, with no sign-up, the AI models’ answers to your market’s questions appear, and the full report follows by email. To compare with the sector, murmur’s sector-by-sector studies list the brands the AI models indicated in each market measured.
Read also
- What is AI share of voice, brand presence in answers
- Reviews and Reclame Aqui influence what AI says about your company
- How to correct wrong information ChatGPT says about your company: the step by step
- How to build the question list to measure brand presence in ChatGPT
References
- Semrush, help center, brand performance reports. The brand’s favorable and general sentiment in AI conversations, the percentage of mentions of each type and the chart of presence against sentiment. https://www.semrush.com/kb/1595-brand-performance-reports (accessed on 9 November 2026). Semrush sells an SEO tool.
- Peec AI, documentation, sentiment metric. The score from 0 to 100, the common range of 65 to 85 and the positive, neutral and negative categories. https://docs.peec.ai/metrics/brand-metrics/sentiment (accessed on 9 November 2026). Peec AI sells AI monitoring.
- Profound, sentiment analysis page. Whether the AI portrays the brand positively or negatively, the narratives that gain ground and the sources that shape the story. https://www.tryprofound.com/features/answer-engine-insights/sentiment (accessed on 9 November 2026). Profound sells AI monitoring.
- Jan Ehrlinspiel, Peec AI, study on brand names in AI descriptions, August 2026. One-word English names and the list of about 14 thousand words from Warriner and colleagues (2013); the effect of the name appears only where the AI has little information. https://docs.peec.ai/research/brand-name-bias-in-ai-descriptions (accessed on 9 November 2026).
- SparkToro, “NEW Research: AIs are highly inconsistent when recommending brands or products; marketers should take care when tracking AI visibility”, 28 January 2026, with runs from November and December 2025. In two answers from ChatGPT or Google’s AI, the chance of the same list of brands coming back was below 1 in 100. https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/ (accessed on 2 October 2026).
- OpenAI, “Searching the web with ChatGPT”, OpenAI Help Center. ChatGPT collects approximate location from the IP address and may pass it to partner search engines. https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt (accessed on 20 October 2026).
- murmur, Mindo case study. https://murmur.marketing/en/cases/mindo/ (accessed on 6 November 2026).