Why traditional SEO metrics won't tell the full story
If you open Google Search Console and stare at impressions and clicks, you will not find your AI visibility performance there. That data simply does not exist in the tools most teams have been using for years. AI search engines like ChatGPT, Perplexity, and Gemini do not fire a click in your analytics when they cite your brand in a response. They do not show up as a referral source. They read your content, extract what they need, and surface it to users, often without a single measurable touch point in your existing stack.
This creates a real problem. You know AI search matters, you have invested time or budget into improving your visibility, and now your boss wants a number. A return. Proof. And you are sitting there with a gut feeling that things are improving but no spreadsheet to back it up.
The good news: it is absolutely possible to track AI visibility over time and build a credible ROI case. It just requires a different approach, combining new measurement methods with some old-fashioned qualitative research.
Setting a baseline before you do anything else
You cannot prove improvement without a starting point. Before you optimise anything, run a structured baseline audit across the AI engines that matter most to your audience.
Pick 15 to 30 queries that a real customer might ask. Be specific. Not "best running shoes" but "what are the best running shoes for flat feet under £100". These are the kinds of conversational, intent-driven questions that AI engines are built to answer. Write them down in a spreadsheet.
Then, manually query ChatGPT (GPT-4), Perplexity, and Google's AI Overview for each one. Record:
- Whether your brand is mentioned at all
- Whether you are cited as a source (with a link)
- Your position relative to competitors (first mention, second, not mentioned)
- The exact language used to describe your brand or product
Do this on a fixed schedule. Monthly is realistic for most teams. Quarterly is the minimum. The data compounds over time, and consistency matters far more than frequency.
If you want a structured starting point, a free AI visibility audit can help you map where you currently stand before you begin tracking changes.
The metrics that actually matter for AI search
Once your baseline is set, you need a repeatable framework. Here are the metrics worth tracking, broken into two tiers.
Tier 1: Presence metrics
These tell you whether you exist in AI-generated answers at all.
- Brand mention rate: Out of your 20 tracked queries, how many responses include your brand name? Express this as a percentage. If you are in 4 out of 20 responses, your mention rate is 20%.
- Citation rate: Of those mentions, how many include a direct link back to your site? This is distinct from a brand name mention. Perplexity in particular links its sources prominently, so citation rate is especially important there.
- Competitive share of voice: For each query, list every brand mentioned. Calculate what percentage of total brand mentions across all responses belong to you. If there are 60 brand mentions across your 20 tracked queries and your brand appears 12 times, your share of voice is 20%.
Tier 2: Quality metrics
These tell you how well you are being represented, not just whether you appear.
- Sentiment accuracy: Is the AI describing your brand in the way you would want? Is the framing positive, neutral, or negative? Does it accurately reflect your positioning?
- Answer position: When your brand is mentioned, does it appear in the first sentence of the response, the middle, or as an afterthought? First mentions carry more weight with users.
- Schema coverage score: Internally, track how many of your key pages have validated, complete schema markup. This is a leading indicator; structured data improvements typically show up in AI mentions 4 to 8 weeks later.
Connecting AI visibility to business outcomes
Presence metrics are useful, but your boss wants to see revenue impact. Bridging that gap requires some lateral thinking, because the attribution chain between "Perplexity mentioned us" and "customer bought something" is not yet direct in most analytics tools.
Tracking brand search uplift
When AI engines mention your brand in answers, a portion of users will then go to Google and search for you directly. This means brand search volume in Google Search Console is a reasonable downstream proxy for AI visibility. If your brand name queries in GSC increase over the same period your AI mention rate improves, that correlation is worth documenting.
Set up a GSC filter for your exact brand name and track monthly impressions and clicks. Plot this alongside your AI mention rate. Over 3 to 6 months, patterns tend to emerge.
Direct referral traffic from AI platforms
Perplexity does pass referral traffic. In Google Analytics 4, create a segment for traffic from perplexity.ai and you.com. These will not account for all AI-driven visits, but they are measurable. Track sessions, bounce rate, pages per session, and conversion rate for this segment separately. AI-referred visitors often behave differently to organic search visitors, frequently arriving with higher intent because they have already had their question partially answered.
ChatGPT traffic shows up inconsistently, sometimes as direct, sometimes from chat.openai.com. Create a segment for both and monitor it over time.
Assisted conversions and the dark funnel
A growing share of the buying journey now passes through AI tools before a customer ever reaches your site. This is often called the dark funnel. Someone asks ChatGPT what accounting software to buy, gets a recommendation, and then types the brand name directly into Google. That purchase shows up as direct or organic in your analytics, with no trace of the AI interaction.
The most honest way to surface this is through customer surveys. Add one question to your post-purchase or post-signup flow: "How did you first hear about us?" Include "AI assistant or chatbot (e.g. ChatGPT, Perplexity)" as an option. Even a modest response rate will give you data over time. Many businesses are surprised how frequently this option gets selected once they start asking.
Building the ROI report your boss will actually read
Once you have 2 to 3 months of data, you can start structuring a proper report. The format matters as much as the numbers.
Keep it to one page
Seriously. Decision-makers do not want 12 slides of methodology. Put your headline metrics at the top: mention rate, citation rate, share of voice, and brand search volume change. Then show the trend over time. Three data points make a line. Six make a story.
Frame improvements in money where possible
If your brand search volume has increased by 400 queries per month, and your typical conversion rate from branded search is 8%, and your average order value is £120, you can model that as roughly 32 additional conversions worth £3,840 per month. That is not a precise attribution, and you should say so, but it is a credible range that grounds the conversation in commercial reality.
Show the competitive gap closing
If your share of voice in AI responses has moved from 10% to 22% over six months while a named competitor has stayed flat or dropped, that is a story leadership responds to. Frame it as market position, not just a technical metric.
Attribute improvements to specific actions
Your report will be far more credible if you can say "we added FAQ schema to 40 product pages in March, and our citation rate in Perplexity increased from 15% to 31% in April and May." That kind of before-and-after narrative is what separates a technical exercise from a genuine business case. This is exactly the kind of outcome that structured schema work can produce, and it is why teams working with specialists like FlinnSchema track implementation dates carefully alongside their monitoring data.
For more context on the types of schema that drive these results, take a look at our post on common schema markup mistakes that hurt AI visibility, which covers the errors that suppress citations even when brands are doing most things right.
Tools that can help automate the tracking
Manual querying works well for small query sets and tight budgets. As your programme scales, it becomes unsustainable. A few tools are starting to fill this gap.
- Profound: Specifically built for AI answer monitoring. Tracks brand mentions across ChatGPT, Perplexity, and others at scale. Relatively new but growing fast.
- Otterly.AI: Tracks brand visibility in AI-generated search results and provides share-of-voice data across a range of AI engines.
- BrandMentions and similar: Not AI-specific, but useful for tracking when your brand name appears in AI-generated content that gets published on the web.
- Custom Google Sheets setup: For most small to mid-sized e-commerce brands, a well-structured manual tracker updated monthly is still highly effective. The discipline of doing the queries yourself also keeps you close to the actual answers, which has its own value.
The category is moving quickly. Tools that did not exist six months ago are now genuinely useful. Keep an eye on what is available, but do not wait for perfect tooling before you start tracking.
How long before you can show meaningful results?
Expect 6 to 12 weeks before schema and content changes meaningfully shift your AI mention rates. AI models are retrained periodically, and crawlers need time to index your updated markup. Perplexity tends to reflect changes faster than ChatGPT, because it uses live web retrieval rather than a static training dataset.
This is worth communicating clearly to stakeholders upfront. Set the expectation that month one is baseline setting, months two and three are early signal, and months four through six are where trends become defensible. Anyone promising overnight results in AI visibility is not telling you the full picture.
If you want to understand how your site looks to AI crawlers right now, the free AI visibility audit gives you a concrete picture of where the gaps are and what to fix first.
Frequently Asked Questions
Can I track AI visibility without paid tools?
Yes. A consistent manual process using a spreadsheet, a fixed set of queries, and monthly recording across ChatGPT, Perplexity, and Google AI Overviews is entirely viable, especially for brands just starting out. It takes roughly 2 to 3 hours per month to run properly. Paid tools become worthwhile once you are tracking more than 50 queries or need to report at scale.
How do I prove that a schema change caused an improvement in AI mentions?
You cannot prove causation with complete certainty, but you can build a strong correlational case. Record the exact date of every schema implementation. Then track your mention rate in the 4 to 8 weeks that follow. If your rate increases after implementation and was flat before, that is meaningful evidence. Documenting this consistently across multiple implementations strengthens the case further.
What if my boss does not understand what AI visibility is yet?
Start with the customer behaviour angle rather than the technical one. Explain that a growing number of buyers are asking AI tools for product recommendations before they search Google, and that brands which appear in those AI answers win consideration before any other channel gets a look in. Frame it as a new distribution channel, not a technical SEO project. Most business leaders respond to that framing immediately.
Is brand search volume a reliable proxy for AI visibility impact?
It is a useful signal, not a precise measurement. Brand search volume captures people who heard your name somewhere and then searched for you. AI mentions are one source of that, but not the only one. Word of mouth, PR, and social all contribute. That said, if your brand search volume is rising consistently during a period when your AI mention rate is also rising, and you have not run a major PR campaign or paid brand campaign, the correlation is worth including in your reporting as supporting evidence.