AI Visibility Measurement: What To Track & What To Ignore

Traditional SEO metrics are breaking down with the rise of AI answers, prompting marketers to rethink measurement. According to the article, most AI visibility tools like Profound, Peec, and AirOps recommend tracking prompts based on existing pages, but these often miss the prompts that matter. The article advises tracking prompts, AI visibility across models like ChatGPT, Gemini, Google AI Mode, Perplexity, Claude, and Copilot, and self-reported attribution to tie efforts to business outcomes.

I have dozens of conversations per week with folks in growth and marketing, ranging from directors, VPs of marketing, and CMOs to SEOs in the nitty-gritty day-to-day. Many of my conversations involve measurement. This is an increasingly challenging topic as traditional SEO metrics https://www.searchenginejournal.com/how-to-define-report-seo-kpis-that-actually-move-the-c-suite/576233/ are breaking down with the advent of AI answers. Much of the conversation is spent debunking misconceptions and misguided advice operators see on social media. It’s not an easy conversation, but it’s important. This means telling people to avoid many of the things they see people promote on social media. It’s not an easy conversation, but it’s the right conversation. My goal here is to clear up the confusion so you can tie your AI visibility efforts to business outcomes. What To Track These are a mix of leading and lagging indicators that you have varying degrees of control over. I’ll make the case for why each one matters, then we’ll cover how to influence them. Prompts This is the most obvious and most important decision, because what you measure influences behavior. From dozens of conversations, it’s also what many people get wrong. It becomes the first domino in a chain of mismeasurement. None of the other metrics, like citation share, brand mentions, or visibility, matter if you’re tracking the wrong set of prompts. Most AI visibility tools https://www.searchenginejournal.com/your-ai-visibility-tracker-is-quietly-breaking-your-analytics/573000/ like Profound, Peec, and AirOps will automatically recommend prompts to track, but these are rarely what you should focus on. I haven’t confirmed this, but from what I can tell, they analyze your website and map existing pages back to prompts. They assume the pages already on your site are the ones that should be cited or visible in LLM outputs. That might be true. But most companies we speak to say, “We aren’t appearing for the prompts we want to show up for,” which tells me they don’t have the right strategy and thus haven’t published the right pages. AI Visibility You can see how, if you’re tracking the wrong prompts, you’ll measure visibility for the wrong things. From our perspective across dozens of clients, ChatGPT is the most commonly used LLM, but it’s worth tracking how often your brand shows up for your target prompts across ChatGPT, Gemini, Google AI Mode https://www.searchenginejournal.com/googles-ai-mode-what-we-know-what-experts-think/555482/ , Perplexity, Claude, and Copilot as well. This is important because ChatGPT’s user base skews consumer, while Claude’s user base skews business and enterprise https://www.forbes.com/sites/siladityaray/2026/06/16/claude-makes-more-money-per-user-than-market-leader-chatgpt-report-finds/ . If you’re a B2B business, Claude has fewer users, but those users are using Claude at work, which is the context that matters. Google has also made AI mode more prominent https://blog.google/products-and-platforms/products/search/search-io-2026/ , and with Chrome having dominant market share https://gs.statcounter.com/browser-market-share , many people will see AI mode outputs by default. Tracking each model separately lets you see whether a change in visibility or traffic is isolated to one model or happening across all of them. There are also significant differences in how each model searches for and presents information. For example, Claude’s use of Brave https://www.tryprofound.com/blog/what-is-claude-web-search-explained versus ChatGPT’s use of Bing; they have their different methodologies for selecting sources https://www.searchenginejournal.com/how-chatgpt-actually-picks-sources-i-read-the-network-traffic-not-the-outputs/580895/ ; and each model has preferred sources through media partnerships https://www.frac.tl/ai-media-partnerships/ . Self-Reported Attribution For most of the last decade, marketers leaned on clickstream analytics and UTM parameters https://www.searchenginejournal.com/how-to-prove-pr-business-value-with-utm-parameters-ga4/566928/ to tell them where leads came from. That paradigm has been breaking down for a while, and it now shows an even smaller part of the picture. LLMs are zero-click by design. When someone asks ChatGPT, Gemini, or Claude a question, the answer is provided in the chat. They don’t click through. They research in the conversation, then they might do a Google search for your brand directly or type your URL straight in. Your analytics and CRM platforms will log that traffic as “Organic” or “Direct,” but the reality is that an LLM is what led to the person going to your website. Self-reported attribution https://www.searchenginejournal.com/google-went-not-provided-in-2011-and-blinded-us-chatgpt-just-shipped-its-version/582839/ is the simplest yet highest-signal way to get more of the story. We recommend simply asking people, “How did you hear about us?” There are well-known and documented flaws to self-reported attribution mostly human memory and salience of touchpoints , so it won’t be perfect, but it’s the buyer telling you, in their own words, how they found you. If you capture that information through a field on your lead form, you can then track those lead sources in your CRM to pipeline and closed revenue. That’s the ultimate success metric for a marketing channel. One client that added it discovered ~5% of registrations were coming from ChatGPT despite doing zero work around AI visibility. Below is a chart from our internal dashboard. These are the number of people who filled out our consultation form and stated that they found us through an LLM, split by the HubSpot-tagged source. Where possible, HubSpot automatically tags the source as “AI Referrals” red . We found that 80-90% of leads that came in via an AI platform were incorrectly tagged as “organic” or “direct.” What To Monitor, But Not Set As KPIs These metrics tend to be related but not the ultimate goal, and often not what you have control over. Citations This is the strategic piece that people mistakenly view as the success metric. Off-page sources dominate citations at every funnel stage. Our research on citation sources https://beomniscient.com/blog/how-llms-source-brand-information/ found that, even for branded or bottom-funnel queries, 48% of sources were earned media, 30% were commercial content from other sites, and only 22% were from the brand’s own website. So you should be measuring what percentage of attainable mentions you currently occupy across third-party surfaces. If it’s 1%, there’s massive headroom. And remember that just because you’re cited, it doesn’t mean you’re recommended https://www.semrush.com/blog/the-ghost-citations-study/ or presented in the LLM output. This means that, yes, you should look at citations, but not simply to see if your website is cited, but whether your brand is mentioned on the most cited pages. Sentiment I get a lot of questions from marketers who say, “ChatGPT shows our brand in a less favorable light than it does our competitors. How do we improve that?” Influencing market sentiment about a brand is a massive undertaking that no single person can control. There are tactics you could use to influence the sentiment LLMs present, like engaging on relevant Reddit threads https://www.searchenginejournal.com/why-authority-in-online-communities-such-as-reddit-quora-matters/555851/ or producing content that paints your brand in a better light. Ultimately, what LLMs present about your brand is not what the LLM thinks–it’s what the market is saying about your brand across hundreds of websites. That has more to do with people’s experience with your product, your sales team, your customer support, and the overall customer experience your company provides. Marketing is just a small piece of that. No amount of marketing will overcome a negative experience with your brand. So monitor sentiment, but treat it less as something to influence through marketing and more as feedback to improve the customer experience, which in turn improves sentiment. LLM Referral Traffic This is worth monitoring, but not a good KPI because you can’t control this. OpenAI recently changed how ChatGPT presents sources and reduced the number of sources it showed https://www.searchenginejournal.com/chatgpt-search-is-citing-fewer-sites-data-shows/571219/ . As a result, many websites lost ChatGPT referral traffic https://beomniscient.com/blog/why-your-chatgpt-traffic-just-fell-off-a-cliff/ . However, they reversed that change a month later and showed more citations, leading to a large increase in ChatGPT referral traffic. That makes for a volatile KPI. Beyond that, a lot of LLM traffic doesn’t get tagged as LLM referral traffic https://www.searchenginejournal.com/ga4s-ai-assistant-channel-undercounts-your-ai-traffic-how-to-build-one-that-doesnt/580133/ by tools like HubSpot or Google Analytics because the referral source is getting stripped. So traffic ends up being bucketed under direct. How To Measure The Most Important Metrics How Do You Track The Right Prompts? The mistake many teams make is letting their AI visibility tool pick their prompts for them or trying to guess what prompts to track. The problem is that those tools don’t know your customers. The good news: you do. That knowledge already exists as sales call recordings and transcripts, onboarding calls, and customer research calls. These are rich sources you can mine for voice of customer to understand: - What questions come up frequently? - What objections come up? - What language do they use to describe pain points or challenges? If your prospects are asking these questions on a call, they – and people like them – are asking the same questions of ChatGPT or Claude. The other source is self-reported attribution see below . Once you’ve implemented self-reported attribution, your sales team can see whether a lead came through an LLM and simply ask them what prompt they used or even request a screenshot of the chat thread. In my experience, people are happy to share. Between call transcripts and direct buyer input, you replace guesswork with the exact language customers use. Your target prompts can and probably should change. We recommend reviewing prompts against voice-of-customer research quarterly, especially if you’re in a fast-moving industry. How Do You Measure AI Visibility In ChatGPT, Claude, Gemini, Etc.? Once you have your set of prompts, we recommend using Profound, Peec, or AirOps for tracking prompts. Watch for the default channels that get tracked. For example, we don’t care much for Perplexity, and Claude is often not included as a default LLM. So we prioritize Claude, ChatGPT, Gemini, and AI Mode. Put in your prompts, track your competitors, and track your visibility over time. How Do You Set Up And Report On Self-Reported Attribution? I mentioned above that platforms like Google Analytics and HubSpot don’t give the full picture of a lead source. We recommend having a method to capture self-reported attribution so that a person can explicitly tell you how they found you. This won’t be perfect attribution never is , but it will give you another informative data point. This will also allow you to track sales opportunities, pipeline, and ultimately revenue that came in via LLMs. I’ve seen this done in three ways: Add a question for “How did you hear about us?” in your lead or product signup forms. You can keep it open-ended or offer a dropdown with pre-defined options and have “AI Assistant ChatGPT, Claude, Gemini, etc. ” as an option. Some clients are averse to this because they don’t want to negatively impact conversion rates. I’d recommend running an A/B test to see if that’s actually the case. Add the “How did you find out about us?” question to your product onboarding flow. This addresses the concern about conversion rates, but doesn’t account for pure sales-led motions. Have your sales team ask on their discovery calls. This requires a behavior change from your sales team, and the downside is it doesn’t account for self-service products. Focus On What Pays The Bills When we talk about paid marketing, we usually don’t measure success by the number of impressions because impressions don’t pay the bills. Instead, we talk about return on ad spend https://www.searchenginejournal.com/calculate-roas/380373/ . For every dollar we put in, how many dollars do we get out? It isn’t quite apples-to-apples, but the same logic applies. Instead of measuring visibility or citations which don’t pay the bills , we should ask whether we’re reaching the right people through LLMs, and whether that visibility translates into business outcomes. That means focusing on leads, pipeline, and revenue. More Resources: How To Measure AI Search Visibility https://www.searchenginejournal.com/how-to-measure-ai-search-visibility/579893/ AI Visibility Rankings Aren’t Stable – New Research Shows It’s Mostly Statistical Noise https://www.searchenginejournal.com/ai-visibility-rankings-arent-stable-new-research-shows-its-mostly-statistical-noise/581905/ We Need To Change Our Approach To AI Prompt Tracking https://www.searchenginejournal.com/we-need-to-change-our-approach-to-ai-prompt-tracking/579646/ Featured Image: RobinRmD/Shutterstock