.png)
Imagine sitting across from a high-stakes prospect for a big pitch. Then, out of nowhere, they pull up an AI-generated comparison deck on screen that completely misrepresents your product.
It’s every marketer’s worst nightmare (or one of them), and yet, it’s the exact situation Monica Kumar, EVP and Chief Marketing Officer of Extreme Networks, found herself facing.
"We were in front of a big prospect," she shared at the CMO Summit in San Francisco. "They slapped in front of us this huge competitive analysis that they had put together using AI. And lo and behold, we didn't look that favorable. But they had the wrong information."
The prospect did their homework, but the AI behind it fed them completely wrong information, which cost Monica’s team the deal before they could set the record straight.
Unfortunately, this happens more often than you might think. Many buyers meet your brand for the first time through AI, leaving your first impression entirely up to what the algorithm decides to summarize.
Joris Brabants, CMO at Apicbase, recently noted that 35% of their leads discover them through LLM searches (source: CMO Insights Report 2026). That means a third of their entire pipeline finds them through an AI model rather than a website or a sales rep. We can then conclude that whatever that model gets wrong about their brand (or anyone else's), it’s getting it wrong at scale.
What makes this dangerous is that none of us were invited to write that first draft of our own brand story. AI models form an opinion without us, pulling from old press releases, competitor blogs, or stale reviews. And whether that picture is accurate isn't the model's problem; it's ours.
Most leaders don't realize what AI is telling prospects about their business until a buyer brings it up on a call. You don't have to wait to get blindsided. In this guide, we'll walk through how to easily track your AI visibility, catch misinformation, and clean up bad data before it damages your brand’s reputation.
How does AI source information about your brand?
AI scrapes data from your website, news articles, third-party review sites, and public forums like Reddit to summarize what your company does. Basically, AI builds an opinion of your brand from whatever it can retrieve and stitch together.
Anirudh Singla, CEO of Pepper, walked a room of CMOs through exactly how this works:
"LLMs are a lot about retrieving content, and the more easily they can retrieve that content, the better you show up."
The keyword there is retrieve because AI doesn’t pull your webpage in its entirety. It cites a few sentences that best answer the user's prompt.
Another thing to keep in mind is that your site isn’t the only place AI will source information from. As we touched on, AI scans YouTube video descriptions, LinkedIn posts, review sites, and third-party publications to help answer prompts.
That all sounds great until you realize you don’t control any of that information. AI can form an opinion about your brand based on an old Reddit thread, a competitor’s comparison page, or a dated review buried online somewhere. All of that information, along with what you’ve actually published, is how AI comes to a conclusion to share with its users.
A quick rundown of how AI sources information:
- Baseline memory and live searches: LLMs build their core knowledge from massive training datasets, and many will run live web searches using crawlers to grab fresh info.
- Your third-party digital footprint: AI looks way beyond your official site and will gather info from all corners of the internet. Remember, AI loves repetition. So, whatever claim gets repeated the most across the web usually becomes the story it tells, even if that story is outdated or wrong.
- Easy-to-read code and content: Models love clear, well-structured information. Using code like JSON-LD schema helps AI map out your products, team, and services without guessing.
The cleaner your site structure (think plain-English FAQs and clear bullet points), the easier it is for the model to pull accurate facts about you.
Alexander Ruban
How to track what AI says about your brand
Keeping AI honest about your brand isn't as hard as it might seem. You just need a simple process to track what models are saying and the right sources to update when they get it wrong. To do that, try the following steps:
1. Build your question/prompt list
A question/prompt list is a fixed set of real buyer questions that you repeatedly run through ChatGPT, Claude, Perplexity, and Gemini. Having one of these lists ready to go is useful because it gives you a good baseline to compare against over time or across competitors.
Anirudh Singla laid this out plainly during his talk at the CMO Summit in Austin 2026, when he shared:
"My first recommendation is going to be to put together a list of a hundred to two hundred queries that you care about, and your buyers care about.
"Pull up information from your Gong, from your CRM, and ask your AEs or sales reps what people are really asking about. For consumer companies, pull up information from your customer reviews, your customer sentiments."
Don't guess at what people are asking; go find out. You should also make sure a chunk of that list names competitors such as "X vs Y," or "best alternative to X for [use case]". These types of prompts matter because that's where reputation risk lives, not in generic "what is [category]" queries.
How to structure your prompt list
Alright, so you know you need to create quite a substantial prompt list, but what does that look like in reality?
One great way to approach it is by thinking like your ideal customer. Then, map your prompts to how customers research, compare, and decide on a solution. Something like this:
Notice that two of these four stages (comparison and brand reputation) are exactly where Monica’s ambush happened. That's not a coincidence. It's where the competitive risk concentrates, which is why they deserve more of your query list than "what is" questions ever will.
2. Run the list across every AI model your buyers use
You’ve got the list, now it’s time to do some good old-fashioned manual labour… sort of. The next step is to run your list across every model your customers use and log what comes back.
Even with their similarities, ChatGPT, Claude, Perplexity, and Gemini don’t agree on everything. One might get your brand right, while another could be working off out-of-date information that’s completely irrelevant today. That disagreement is itself useful data, but only if you capture it.
A simple spreadsheet works fine. Just create one row for each query, and add these columns:
- Date
- AI Model (ChatGPT, Perplexity, Gemini, etc.)
- Brand Mentioned? (Yes / No)
- Competitor Mentioned? (Yes / No)
- Accuracy (Correct / Incorrect / Partial)
- Cited Sources (Links to the sources the AI pulled from)
Since these tools are non-deterministic, it means that the same prompt can return a completely different answer just minutes apart. So, a single run of a query can fool you into thinking something’s fixed or broken when it isn’t.
To help retrieve better data for your tracking system, run each query two or three times per model. This will help you build a far more accurate 'average' score for your brand.
Tips for this step:
- Categorize responses and take note if your brand was just mentioned (named in passing), recommended (pitched as a top pick), or cited (linked directly as a source).
- Carry out this step logged out, in a fresh chat each time, so you're seeing something closer to what an anonymous buyer sees rather than a result shaped by your own account history.
- Run each query for your brand and your named competitors. You’ll often discover that the useful signal isn't "are we mentioned," but "who's getting mentioned instead of us."
3. Read the visibility and citation columns together
Once you’ve filled out your tracking spreadsheet, pay close attention to the gap between your columns.
High visibility paired with low citations is the ultimate danger zone. It means the AI is talking about your brand, but it isn't actually reading or quoting your website to do it. If this is the case, you’ve got no say in how your brand is described.
If you spot this pattern in your data, don't panic and start churning out dozens of new blog posts. The fix isn't creating more content. Instead, you’ve got to improve and tweak the content you already have so that it’s easier for AI models to cite.
4. Check your web logs, not just Google Analytics.
This one gets missed constantly. AI crawlers like GPTBot, PerplexityBot, ClaudeBot, and Google-Extended don't show up in Google Analytics. But they do show up in server logs.
Chris Andrew, CEO of Scrunch, opened this part of his New York talk at the CMO Summit with a show of hands:
"How many people have an active way to monitor your presence across these models? How much traffic are you getting from them? Any hands in the air?"
Out of a packed room, two people raised their hands. Two. But we can’t say we’re all that surprised since it lines up with a wider pattern we uncovered in our CMO Insights report 2026.
When marketing leaders were asked to rate their confidence in the data behind their own attribution, the average came out to 2.8 out of 5, the lowest score anywhere in that entire study. Most brands genuinely don't know what's happening here, but Chris shared how to get around the issue:
"First thing I would encourage brands to do is go look at the data. Don't estimate volumes. Don't go try to guess how much stuff is happening.
“You can go to your web logs, talk to your IT team, and find out how frequently these AI agents and crawlers are hitting your website. You're not gonna see it in GA, you're gonna see it in your web logs."
As we touched on, AI crawlers don't register in Google Analytics, but they do show up in server logs. 10 minutes with whoever owns your web infrastructure will tell you more than a month of guesswork.
5. Audit and log these metrics
Some useful metrics will help you track what AI is saying about your brand. We recommend you pay close attention to the following:
Citation rate & rank
To track citation rate, note whether the AI included you at all (your citation rate) and where you landed on the list (your recommendation rank).
Those numbers matter because AI users hardly ever read past the first two or three suggestions. So, being listed at the top frames your brand as the primary authority. If your brand isn’t included in the top three spots, the chances of a user actually clicking through to your site drop significantly.
A text mention means the AI dropped your name, but a citation means it included a direct link to your site. Higher rank directly correlates with getting actual clickable citations.
To log this, simply check your query outputs and note your exact position (e.g., "Mentioned 3rd out of 5, cited with link").
AI Share of Voice
AI Share of Voice measures your brand’s total presence in AI-generated answers compared to your direct competitors across a set of target queries.
Tracking your SOV shows whether your brand is being included in those limited recommendation spots. Monitoring this percentage over time reveals if your visibility is increasing or if competitors are gaining more mentions in your category.
To calculate your score, count every brand mentioned across all your test results. Then, divide your brand's total mentions by the total number of all brand mentions combined.
For example, if your test prompts generate 50 total brand mentions across all outputs and your brand is named 12 times, your AI Share of Voice is 24%. This gives you a concrete percentage showing your exact market representation in AI outputs.
Sentiment
It’s not enough to simply check if your brand appears or not. You’ve also got to look at how the AI frames your brand. Is it an enthusiastic endorsement, a hesitant "yes, but...", or an outright ‘lie’ from an old forum thread?
Pay attention to specific word choices because they reveal the AI's underlying logic:
- Authority indicators ("industry standard," "popular," "top-rated"): These tell you the AI considers your brand a primary market leader based on high search frequency and trusted review signals.
- Hesitation markers ("however," "but," "caveat"): Words like "great features, however expensive..." show the AI is pulling negative sentiment from pricing pages or competitor comparisons, which can stall buyers in the consideration phase.
- Outdated tags ("lacks feature X," "known for downtime"): If the AI repeats an issue you fixed two years ago, it means its live search is over-weighting old Reddit threads or stale third-party reviews over your official site.
"Analyze the sentiment of your previous response about [Brand Name]. List the exact pros, cons, and qualifier words you used, and explain what sources drove those conclusions."
Source provenance
Where is the AI actually pulling its facts when it talks about you? Is it citing your official homepage, an unbiased review platform like G2, a thread on Reddit, or a 3-year-old news article?
Tracking where those links point over time reveals which third-party sites hold the most weight in your category.
Engines like Perplexity, Gemini, and ChatGPT Search list footnotes and direct source links right inside their answers. For models that don't output clickable links, follow up with a direct prompt: "What specific sources, websites, or documents did you use to generate that answer?" Log those domains in your audit sheet alongside your brand's ranking.
Anastasia Glavatchi
AI monitoring tools
Manual tracking gets you a long way, but at some point running spreadsheets against a dozen competitors and a couple hundred queries every quarter turns into a full-time job.
Thankfully, there are some AI monitoring tools you can use to do the brunt of the work for you. Some examples include Pepper’s Atlas, BlueOcean AI, Search Atlas, and more. This space is moving fast, so we recommend checking out a few demos before settling on one.
Fixing what AI gets wrong
When AI gets your brand wrong, there's no support ticket you can submit to fix it. You have to overwrite the narrative yourself. Here’s how to do it.
1. Fix your own content first
82% of AI search citations come from third-party sources rather than your official domain. This happens largely because models ditch official websites the moment they spot contradictory information.
For example, if an archived FAQ on your site clashes with your main homepage, the AI gets confused and defaults to outside reviews instead. So, make sure your own content doesn’t contradict itself.
Audit your URLs, purge legacy product pages, and lock down your facts. Eliminating mixed signals on your own domain is the fastest way to build the trust AI needs to cite you correctly.
2. Trace the bad source when it's not you
Models rarely invent false claims out of thin air. But they will copy them from somewhere else on the internet.
When you see AI making a mistake about your brand, go ahead and ask the tool directly: “What specific sources did you use to answer that?" Tools like Perplexity, Gemini, and ChatGPT Search will hand you a list of links.
Once you find the root cause, you have two options:
- Update the source: Reach out to the publication or platform to correct the outdated data directly.
- Out-publish it: If it's an uneditable forum post, publish newer, stronger, and better-structured content on authoritative sites the model already trusts.
3. Publish a fact-correction page in BLUF format
AI search engines love the Bottom Line Up Front (BLUF) content. They evaluate text paragraph by paragraph, looking for standalone, authoritative answers to pull into an answer.
If an engine keeps insisting your platform lacks a key feature, don't write a fluffy story about it. Publish a dedicated FAQ or correction page and get straight to the point.
- Ditch the fluff: "We've always prioritized innovation, which is why last quarter our engineering team rolled out..."
- Use BLUF: "Yes, [Brand] includes full [Feature] support across all enterprise plans as of 2026."
Keep those first 30 to 50 words completely self-contained so the crawler can lift them cleanly.
4. Feed the machines structured JSON-LD schema
LLMs hate ambiguity. If key details like your features or founder history only exist in decorative site copy, the engine has to interpret your text. The problem then becomes whether that interpretation is correct or not. In many cases, it leads to errors that need to be corrected.
JSON-LD schema markup is code-level metadata that acts like a direct cheat sheet for AI crawlers. It explicitly defines:
- Your exact company name (and common aliases)
- Current product names and active pricing tiers
- Core features and service offerings
- Verified links to your social profiles and review listings
Think of schema as a direct API feed for AI bots. It eliminates the translation layer and removes the guesswork entirely.
5. Unblock the AI crawlers
You can write the most accurate content in the world, but if your technical setup blocks AI search bots, the models default back to old third-party chatter instead.
Check your robots.txt file and server settings (Cloudflare especially) to make sure you aren't accidentally disallowing GPTBot (OpenAI), PerplexityBot (Perplexity), ClaudeBot (Anthropic), or Google-Extended (Gemini / AI Overviews).
If those bots are blocked in your code, the model literally can’t read your updates, and no amount of great content fixes that.
6. Build third-party consensus (the 80/20 rule of GEO)
Models prioritize consensus over what you claim on your own site. So, if your website says one thing and 10 industry blogs say another, you can bet the AI will side with the majority.
To fix a persistent mistake for good, refresh your profiles on Trustpilot and other review sites. You should also pitch updated product facts to industry publications, and engage in Reddit and LinkedIn discussions where live crawlers look for social proof.
Chris Andrew, CEO of Scrunch, watched this work directly: one of his customers mapped the top ten third-party sources most frequently cited in their category, then reached out to build relationships with each one.
"We had a customer that basically looked at the top ten third-party sources that were most frequently cited and reached out and formed relationships with those third-party content providers and said, ' Hey, this is my brand. Whether it's pay-to-play or it's friendly, let's get me on these sources.’ And they would quadruple their AI search traffic by playing nicely with the third parties that are heavily prioritized by the models."
When your site, your review profiles, and the third-party sources the model already trusts all say the same thing, it has no version of the story left to disagree with.
Don’t let AI write your story for you
Monica Kumar didn't lose that deal because her product was weaker. She lost it because an AI model had already told the story wrong, and nobody caught it in time.
That's the real takeaway here. Your brand already has a reputation living inside ChatGPT, Claude, Perplexity, and Gemini right now, whether you've ever checked it or not. The only question is whether you're the one shaping it, or whether you're leaving that job to a third-party source.
FAQs
How can I track my AI visibility?
To track AI visibility, take the questions your buyers actually ask and run them through your AI model of choice. Make sure you’re logged out of your account, though, and run each question a few times each since these tools aren't consistent from one run to the next.
For every answer, note two things: did your brand get named (visibility), and is your own content actually the source (citation). Start with a spreadsheet. Once you're doing this across dozens of competitors and a couple hundred queries, that's your cue to graduate to a dedicated tracker.
How do you track brand awareness?
Track traditional metrics like branded search volume, direct website traffic, and share of voice, alongside your AI visibility score. Combining web search trends with prompt audits gives you a complete picture of modern brand presence.
Why should I track AI brand visibility?
Because your buyers are already using AI to form opinions about you before they ever talk to your sales team. If you're not checking what AI says about you, you're blindly trusting a robot to control your brand’s reputation.
How do you track AEO and GEO?
Monitor your citation share across generative answers, track schema validation on core pages, and map the top third-party sources (like review platforms or media outlets) that AI search engines draw from when summarizing your category.
How does AI source its information?
AI engines use Retrieval-Augmented Generation (RAG) to scan the web in real time, combined with their pre-trained models. They crawl high-authority domains, review sites, community forums, and structured schema data to assemble summaries.
How is AI used in branding?
Beyond content creation and visual asset generation, AI now acts as the primary "discovery engine" for your brand narrative. Managing how models perceive, categorize, and describe your business is the newest frontier of brand management.
How does AI choose its sources?
AI search engines prioritize consensus, freshness, and structural clarity. They lean heavily on independent third-party sites, clear Schema.org markup, and pages formatted with direct, plain-English answers to specific queries.
How can AI be used for reputation management?
Use AI tools to continuously monitor brand sentiment across news and social channels, while using Generative Engine Optimization (GEO) tactics to overwrite false narratives and secure accurate citations across major LLMs.
Become a CMO Insider
Thank you for subscribing
Get exclusive insights, frameworks, and strategies from marketing leaders driving real business impact.
An email has been successfully sent to confirm your subscription.
Follow us on LinkedIn
