I’ve said this before and I’ll say it again: just because ChatGPT mentions your brand doesn’t mean it’s recommending you.
Suppose someone asks ChatGPT or Gemini to compare three SEO agencies. Your agency is mentioned in the comparison but your competitor is described as the market leader and you’re just the “cheaper alternative”.
Both got visibility. So who do you think the customer is going to go with?
That’s why I think tracking brand sentiment in the AI space is just as important as tracking visibility.
I don’t just want to know whether ChatGPT, Gemini or Perplexity mention my brand. I want to know what they’re saying about it, why they’re recommending my competitors and what’s driving those recommendations.
AI visibility tells me whether my brand is being seen. Brand sentiment tells me how that visibility might impact a customer’s decision.
What is Brand Sentiment in LLMs?
Brand sentiment in LLMs is about how positive, negative or neutral a company is described in the AI’s output.
It’s not just about counting how many times your brand appears. Sentiment analysis looks at the language, the context and the attributes of the language.
For example, imagine asking ChatGPT to compare three software providers.
| Brand | AI-Generated Description | Sentiment |
| Brand A | A reliable platform with extensive integrations and strong enterprise capabilities. | Positive |
| Brand B | Offers standard project management and collaboration features. | Neutral |
| Brand C | An affordable option, although its advanced reporting capabilities are limited. | Mixed |
These are all hypothetical examples, but they do illustrate the point about sentiment.
All three brands got visibility, their descriptions however convey quite different degrees of confidence and appropriateness. That’s also why I treat sentiment as a separate entity from the other AI visibility metrics.
- Brand mentions are a measure of how often a brand appears in the AI generated responses.
- AI Share of Voice (SOV) is a measure of a brand’s share of mentions as compared to its tracked competitors.
- Citations are used to identify the sources cited or linked to in AI generated answers.
- Brand sentiment looks at whether the mentions are positive, negative or neutral.
A business can have a very strong AI Share of Voice and yet be repeatedly described in a negative way. That’s something that traditional keyword ranking reports simply won’t reveal.
Why Brand Sentiment Matters in AI Search
Search behaviour is changing and I’ve been talking about that quite a bit.
Rather than going to a number of websites to compare products and services, users can simply ask their AI assistant to do the comparing for them, explain the differences and then pick the provider for them. That in turn means that businesses have another place in which their reputation and positioning can impact the customer.
AI Responses Can Shape Brand Perception
Let’s say someone asks ChatGPT:
“Which SEO agencies in the Singapore are the best for enterprise businesses?”
The AI response may suggest a number of agencies, each with their own particular specialism.
If your company is in the business of being technically proficient, of having the enterprise experience and of having actually delivered on projects then that’s the sort of thing that will go a long way to building credibility with a prospective customer.
But what if AI is repeatedly describing your company as being suitable only for small businesses? Even if that information is incorrect it could in fact put enterprise customers off.
This is why I think businesses need to look beyond the question of whether they’re appearing in AI generated answers. They need to look at whether those answers are in any way communicating their intended positioning.
Positive Mentions Aren’t Always Strong Recommendations
Think about these descriptions:
- “Brand X provide SEO and digital marketing services’
- “Brand X is worth looking at for businesses who need technical SEO and large scale site optimisation.”
Both are good, though the second one explains why you’d go with the company.
This is where the strength of your recommendation and your brand associations come into it. A neutral description may be perfectly accurate but will do nothing to convey what makes your business stand out. A positive description may highlight strengths that are in fact not of any real relevance to your customer base.
What matters is not just getting the positive mentions. It’s being recognised for the qualities that make your business the one you’d choose.

How Do LLMs Form Opinions About Brands?
Before we go any further and start tracking sentiment, we need to be clear about where these descriptions actually come from.
LLMs have no personal views on businesses. Their responses are derived from the patterns they’ve learned during training and where possible from information retrieved from the web.
That information may include:
- Company websites and service pages
- Customer reviews and testimonials
- Industry publications and news coverage
- Business directories and comparison websites
- Expert commentary and research
- Public discussions and third-party references
For instance if a number of credible sources have the company linked to enterprise SEO then the AI systems will have more to go on in that respect.
The other way around is also possible. The business may have expanded its range of services, improved its processes, sorted out the various complaints that were previously being received, etc. but the old information may still be the one that’s being used by some of the AI platforms to describe it.
This is why I think fresh content and AI visibility are so important when you’re assessing how a company is being portrayed.
Publishing updated content doesn’t in any way guarantee that the LLM will instantly reflect those changes. The aim is simply to make accurate, reliable and up to date information about your business more discoverable and more easily referenceable.
How to Track Brand Sentiment in LLMs
I do sentiment tracking much the same as an SEO audit really. You can’t make any improvements without first having a clear understanding of where you are at.
That’s about working out which AI platforms mention the brand, what they say about it and whether the way they describe it varies from platform to platform.
Here’s how you can get going.
Step 1: Identify the AI Platforms to Monitor
Start with the AI platforms your potential customers are likely to be using.
These may include:
- ChatGPT
- Google Gemini
- Perplexity
- Google AI Mode
- Google AI Overviews
I’d recommend looking at those platforms individually as the same question can yield quite different answers.
For example ChatGPT might be very positive about your company whereas Gemini is far more neutral. That doesn’t necessarily mean one is wrong of course. It just means it’s worth being aware of the differences in how your brand is being portrayed.
Also remember that AI generated responses can vary depending on the model, the search context, the user’s location and all sorts of other things.
This is why depending on a single response is not sufficient.
Step 2: Build a Consistent List of Prompts
One mistake I make when assessing AI visibility is to only test questions that are likely to produce a positive answer.
‘Why is Brand X the best?” isn’t a good way to gauge brand sentiment.
Instead I’d recommend creating a fairly balanced list of prompts that reflect how a real customer would go about researching your business.
| Prompt Category | Example |
| Brand Reputation | Is [Brand] a reputable company? |
| Brand Positioning | What is [Brand] known for? |
| Strengths and Weaknesses | What are the pros and cons of [Brand]? |
| Competitor Comparison | How does [Brand] compare with [Competitor]? |
| Category Recommendations | Which companies are best for [service]? |
| Customer Suitability | Is [Brand] suitable for [specific need]? |
Don’t restrict yourself to branded prompts.
Someone asking “What are the best SEO agencies in the Singapore?” may well not have any particular agency in mind. If your company comes up in that answer then how it’s described will be a big factor in whether or not the person goes on to consider your services.
I’d go for the non-branded, commercially relevant questions as those are the ones that will tell you how the AI is positioning your business prior to a customer even searching for your brand.
Keep a consistent core set of prompts so you can make proper comparisons over time.
Step 3: Record and Classify AI Responses
When it comes to manual tracking I’ll use Google Sheets as that’s quite sufficient to get going.
I recommend recording the following information:
- AI platform and date
- Prompt used
- Full AI-generated response
- Brand mentions and sentiment classification
- Positive and negative brand attributes
- Competitors mentioned
- Supporting citations or sources
You can then go and classify each relevant mention into one of four categories.
- Positive: The response clearly favours the brand and/or highlights its strengths.
- Neutral: the response is purely factual with no clear value judgement.
- Negative: The response highlights the negative aspects or criticisms.
- Mixed: the response contains both positive and negative comments.
I actually prefer to split mixed sentiment from neutral sentiment as they’re saying different things.
For instance, describing a software platform as being affordable yet lacking in the more advanced features is far from neutral. It’s clearly stating both the good and the bad.
One important reminder: Always preserve the original AI-generated response.
A sentiment label alone won’t tell you whether the information is accurate, whether the criticism is justified, or what the user was actually asking. You’ll need that context when deciding what to improve.
Step 4: Calculate Your Sentiment Distribution
Once the responses have been classified work out the % of mentions in each category. Here’s how I compute for positive sentiment:
Positive Sentiment Rate = Positive Mentions ÷ Total Classified Mentions × 100
For example, suppose you’ve collected 100 brand mentions across your monitored prompts.
| Sentiment | Mentions | Percentage |
| Positive | 55 | 55% |
| Neutral | 25 | 25% |
| Mixed | 15 | 15% |
| Negative | 5 | 5% |
In this example 55% of the classified mentions were positive.
It’s not a universal reputation score at all. It’s not intended to be a measure of your company’s overall reputation. It only relates to the prompts, platforms and responses that were actually analysed.
For more meaningful insight I’d recommend looking at sentiment by platform, by prompt category and by competitor.
For instance you might find that your brand is getting the ‘affordable’ kudos in the reviews but a rather mixed bag when it comes to the enterprise capabilities. That’s a lot more useful than just having a positive sentiment % figure.
How to Track Brand Sentiment Using AI Visibility Tools
Manual tracking is fine for smaller datasets. As you start looking at hundreds of queries across multiple platforms and competitors though, the process of collecting and classifying everything manually becomes a massive time sink.
This is where AI visibility monitoring tools come in useful. I tend to use our own tool RankAnchor for this.
RankAnchor for AI Sentiment and Share of Voice Tracking
RankAnchor combines AI visibility monitoring with traditional SEO research and optimisation.
One of its features is AI Mention Sentiment which will tell you whether a brand is being mentioned positively, neutrally or negatively in the AI response.
Rather than evaluating sentiment in isolation, I can use RankAnchor to examine it alongside other AI visibility metrics, including:
- Positive, neutral, and negative brand mentions
- AI Share of Voice against competitors
- Visibility across tracked prompts
- Brand mentions and citation opportunities
- Changes in competitive visibility
This is useful as greater AI visibility doesn’t necessarily equate to a more positive brand representation.
For example, imagine your Share of Voice goes up significantly. At first sight that seems like a good thing. What if those extra mentions are all about the shortcomings of your products and services?
Your visibility has improved of course, but that doesn’t necessarily equate to any kind of improved brand positioning. That’s why I’d always recommend tracking sentiment as well as Share of Voice rather than treating them as two separate metrics.
The objective is to understand not just whether your business is getting visibility, but how it’s being represented as that visibility increases.
How to Improve Negative Brand Sentiment in LLMs
Now here’s where I think businesses need to be careful.
Finding negative sentiment doesn’t mean you’ve got to go and rewrite your website straight away.
First of all, check that the information is correct. A valid criticism would require a different response to an out of date or incorrect description.
Correct Outdated or Inaccurate Information
If an AI platform talks about legacy services, wrong locations or outmoded constraints then you need to check the sources that back those up.
Start off by having a look at your own website to ensure that your service pages, your company pages, your product pages etc all accurately reflect what you actually do.
Then check out the relevant third party sources. If a third party publication has some out of date information about your business then you may well want to ask for a factual correction.
Same with any content update really there’s no guarantee that the AI responses are going to change overnight. The benefit of having better, more accurate and more consistent information is that it gives the systems better stuff to work with.
Strengthen Evidence of Expertise
If your competitors are getting better descriptions in terms of experience or reliability then look to see whether your own site has sufficient evidence to back up those claims.
It’s easy enough to describe your business as being at the forefront of the industry. Proving that is a different matter entirely.
I’d recommend beefing up the content that shows off your abilities with case studies, certs, original research, relevant project work and legitimate testimonials.
For example rather than just going on and on about the fact that you’re an enterprise SEO agency, actually publish some evidence of the scale and complexity of the work you’ve done. That gives the customer something tangible to assess and something that an AI might be able to look at.
Build Credible Third-Party References
Your site isn’t the only source of info on your brand.
Independent reviews, industry publications, the various directories and expert contributions can all provide context for your business. That’s where digital PR and off page SEO can help with your AEO and GEO.
We wouldn’t recommend the approach of just going out and publishing generic articles across all sorts of websites in the hope of getting more brand mentions.
As I noted in my piece on Google’s August 2026 spam update, simply producing more generic content isn’t the solution. The content still has to be useful, original and trustworthy.
Address Legitimate Criticism
Not every negative description is an SEO problem.
If customers are consistently flagging problems with your products and/or services then churning out more positive content is of no use whatsoever. The business has to deal with the issues in the first place.
Brand sentiment tracking should be about highlighting the weaknesses and inaccuracies, not about hiding the legitimate criticism.
Occasionally the most useful thing you can learn from a negative AI response isn’t what you need to change on your site. It’s what you need to improve in your business.
How to Measure Changes in Brand Sentiment
Tracking sentiment gives you a baseline. Tracking it over time enables you to see if your brand’s positioning is in fact changing.
I’d recommend looking at your core set of prompts on a monthly basis. For more competitive sectors you may want to be looking at them on a weekly basis to pick up any changes sooner.
The metrics I would prioritize include:
- Positive and negative sentiment percentages
- Recurring brand attributes
- Recommendation presence
- Competitor sentiment differences
- AI Share of Voice
- Accuracy of brand descriptions
I’d also be documenting any changes made to your site, your content strategy, any third party references etc. That way we can establish whether there’s a correlation between any shift in sentiment and your optimisation activities.
However remember that AI responses can vary from run to run. Model updates, retrieval changes, prompt variations and so on can all have an effect. That’s why I’m looking at the patterns over time rather than treating a single positive or negative response as definitive proof that anything has worked.
And when you’re looking at performance over time make sure you’re using the same prompts, the same platforms and the same classification methods otherwise you’re going to be comparing two very different datasets.
Common Mistakes When Tracking LLM Brand Sentiment
The vast majority of sentiment tracking problems stem from the fact that you collect data without any context. The mistakes I would avoid include:
- Treating every brand mention as positive
- Using only branded or leading prompts
- Confusing neutral sentiment with negative sentiment
- Ignoring mixed descriptions
- Comparing different tools’ proprietary scores directly
- Reacting to one unfavorable AI response
- Changing tracked prompts every reporting period
- Assuming more positive content automatically improves AI sentiment
And perhaps the most fundamental error is to think that if you just publish more positive content that it will automatically improve the way your business is described by the AI.
It doesn’t work like that.
The objective isn’t to generate a positive sentiment score. It’s to determine whether the AI is accurately reflecting your company and whether that reflection is in any way going to be of benefit to the customer base you’re trying to reach.
Key Takeaway
Getting mentioned by ChatGPT or some other AI platform is one thing. Being recommended for the right reasons is quite another. That’s why I think brand sentiment tracking should be a core part of every business’s SEO, AEO and GEO strategy.
Don’t just look at frequency of appearance. Look at how AI describes your business, how that compares to your competitors and whether it’s actually describing you in the way you’d want to be described. That gives you the opportunity to go in and correct the record, to bolster your credibility and to improve the way your business is represented in AI.
Because at the end of the day, AI visibility isn’t just about getting mentioned. It’s about making those mentions count.

Sandamalee is a dedicated digital marketing professional and a core member of the Sotavento Medios team. I work closely with my colleagues to manage daily digital operations, ensuring that every project meets the high standards our clients expect. My focus is on maintaining the technical and editorial integrity of our diverse business portfolios, helping brands stay consistent and impactful in their online presence.

