Key takeaways
- A single blended visibility score is misleading. Search Engine Land's prompt-mapping work shows a brand can appear in 70% of branded and comparison prompts but just 10% of problem-discovery prompts, and the aggregate "40% visibility" tells you nothing about either problem.
- Each stage needs its own metric: mention rate for awareness, share of voice and citation-vs-recommendation rate for consideration, and first-mention rate plus conversion data for decision.
- AI engines have different measurement quirks. ChatGPT cites roughly 5 sources per response, AI Overviews about 10, and Copilot's count swings wildly, so cross-engine comparisons need monthly trends, not weekly snapshots.
- Decision-stage measurement lives in your analytics, not just in visibility tools. ChatGPT referral traffic converts at 15.9% versus 1.76% for Google organic (Seer Interactive), so a small GA4 segment is worth building properly.
- Product pages are now the most-cited content type in both ChatGPT and AI Overviews, which means your commercial pages are legitimate visibility targets, not just your blog.
Why one blended visibility score misleads
Most teams I see tracking AI visibility do something like this: pick 100 prompts, run them weekly, report an average visibility score. The number goes up, everyone is happy.
The problem is what the average hides. Casey Nifong's prompt-mapping framework at Search Engine Land describes this exact failure: a brand tracking 100 prompts at an aggregate 40% visibility might actually appear in 70% of branded and comparison prompts but only 10% of problem-discovery prompts. You're winning the shortlist and losing the buyer two steps earlier, and the blended score shows none of it.
There's a second problem with averages: equal weighting. If you combine 50 broad awareness prompts with 10 high-intent evaluation prompts into one visibility rate, the awareness prompts dominate the math and, in Nifong's words, obscure what's happening closer to purchase. Ten decision-stage prompts where you're absent are worth more than fifty awareness prompts where you appear once.
So the fix is structural, not technical. Tag every prompt by funnel stage before you start measuring, and report each stage separately.
Step 1: Build a stage-tagged prompt map
Prompt mapping is the AI-search version of keyword mapping: list the questions your buyers actually ask AI platforms, then organize them by topic, intent, persona, and funnel stage. At minimum, tag every prompt by stage and core topic. Persona, geography, and product line tags are useful later for pattern analysis, but stage is the non-negotiable one.
Here's what the stages look like in practice:
| Stage | Buyer mindset | Example prompts | Primary metric |
|---|---|---|---|
| Awareness | "I have a problem" | "Why is our churn so high?" / "How do B2B SaaS teams track pipeline leakage?" | Mention rate, unlinked brand mentions |
| Consideration | "What kinds of solutions exist?" | "Best churn prediction software for mid-market SaaS" / "How to identify at-risk customers before they cancel" | Share of voice, citation vs. recommendation rate |
| Decision | "Which one should I buy?" | "[Your brand] vs. [Competitor]" / "Is [Your brand] worth it for a 50-person sales team?" | First-mention rate, recommendation rate, conversion |
A note on prompt volume. Semrush data cited by Search Engine Land puts 65-85% of AI prompts outside any traditional keyword database, so don't just port your keyword list over. Build prompts the way people actually talk to AI: long, contextual, and specific. "Best CRM software" becomes "What CRM is best for a 50-person B2B sales team that uses HubSpot for marketing and needs better pipeline reporting?" Change the company size, stack, or pain point and you have another plausible prompt.
How many prompts? Industry guidance converges on 50-100 to start. If you're tracking your own brand, weight the set toward the stages closest to purchase: light on awareness, heavier on decision. Even distribution makes more sense for industry-wide research. And because only 30% of brands stay visible from one AI answer to the next (AirOps' 2026 State of AI Search report), run the set repeatedly. A single manual snapshot is close to worthless.
Step 2: Pick the right metric for each stage
AirOps recommends a clean split: mention rate for awareness, share of voice for consideration, first-mention rate for decision. I'd add one refinement, which is separating citations from recommendations at the consideration and decision stages. Visibility Labs found that brands whose own listicle content was cited were still excluded from the actual recommendation 69% of the time. Being a source is not the same as being the answer.
Awareness: are you part of the problem space?
Awareness prompts center on problems, symptoms, goals, and educational questions. The buyer doesn't know your category exists yet, and may not know your category's name.
Measure two things:
- Mention rate: the percentage of awareness prompts where your brand appears anywhere in the response, linked or not. Brands that earn both citations and unlinked mentions are 40% more likely to resurface across multiple AI answers than citation-only brands, so track mentions separately from links.
- Association quality: when you do appear, is it in the right context? If you sell churn prediction software but only surface when people ask about "customer success tools" generically, the AI has filed you in the wrong drawer.
The honest caveat about awareness measurement: most AI answers carry no clickable citation link, so click-based analytics systematically understate awareness-stage influence. A buyer can read your brand name in a ChatGPT answer, remember it, and Google you a week later. Your GA4 will call that "direct." A "How did you hear about us?" field on lead forms is crude but it catches what referral data can't.
Consideration: are you associated with the solution?
Consideration prompts introduce solution categories, capabilities, and approaches. This is where the AI is building the buyer's shortlist.
Measure:
- Share of voice: your share of mentions across all consideration prompts, compared against your named competitors. This is where competitor heatmaps earn their keep.
- Citation rate vs. recommendation rate: track both. If you're cited in 40% of "best X software" prompts but recommended in 10%, your content is being strip-mined for facts while someone else gets the endorsement.
- Sentiment: how the AI describes you when it names you. "Reliable but expensive" is a different problem than "popular with small teams."
One data point worth knowing: Reddit has historically been a big consideration-stage citation source, but Promptwatch's data shows reddit.com's share of ChatGPT Search citations collapsed from roughly 4% to 0.5% on August 14, 2026, an 86% relative drop in a single day following ChatGPT's query-fanout behavior change on August 8. If your consideration-stage strategy was "get mentioned in the right subreddits," that assumption needs a re-check. AI Overviews and AI Mode show only gradual Reddit declines over the same window, so the picture differs sharply by engine.
Decision: do you win the shortlist?
Decision prompts are brand-specific: comparisons, pricing, "is it worth it," and validation questions. This is the stage where measurement gets most concrete because it connects to money.
Measure:
- First-mention rate: when a comparison prompt lists options, are you first? Position matters more here than anywhere else in the funnel.
- Recommendation rate: when the AI picks a winner or narrows to two, are you in it?
- Shopping feature presence: ChatGPT attaches shopping cards to a low single-digit percentage of web-search responses overall, but that concentrates heavily on commercial prompts, and the rate moves in step-changes OpenAI controls directly. Promptwatch's shopping usage data shows it roughly doubled overnight in late May 2026, then fell back weeks later. Run your transactional prompts regularly and log which brands fill the cards, because the feature can be toggled for your category at any time.
- Conversion: the analytics piece, covered next.
One more decision-stage wrinkle: brand accuracy. SOCi's 2026 Local Visibility Index found business profile information on ChatGPT and Perplexity was only 68% accurate. If the AI is confidently telling buyers your pricing, locations, or feature set wrong, that's a decision-stage visibility problem no amount of content fixes.
Step 3: Understand each engine's measurement quirks
Cross-engine visibility numbers are not directly comparable, and knowing why saves you from bad conclusions.
Promptwatch's analysis of sources per response (built on 26B+ analyzed citations, prompts, and responses) found ChatGPT cites about 5 sources per web-search response, while Google AI Overviews and Perplexity both cite around 10. Two consequences:
- ChatGPT's ~5 slots are far more contested than a traditional SERP. Missing a citation there means something different than missing one in AI Overviews.
- Perplexity's citation count is remarkably stable day to day, which makes it the best "control" engine when you're testing whether a content change moved the needle. Microsoft Copilot, by contrast, has swung from under 2 to nearly 17 sources per response within weeks. Judge Copilot on monthly trends, not weekly snapshots.
Query fanouts matter too. ChatGPT breaks a single prompt into multiple web searches, each targeting a different angle, so one awareness prompt can generate several distinct retrieval opportunities. But Promptwatch's fanout data shows average fanouts fell from 2.15 in December 2025 to 1.0 by April 2026, and average fanout query length dropped from ~117 characters to ~53. ChatGPT is searching in terse, keyword-style bursts now, which favors pages organized around one specific sub-question over broad catch-all pages. Then on August 8, 2026, ChatGPT Search started using the site: operator at scale, jumping from ~0.4% to ~17% of fanout queries overnight, with searches per response nearly doubling. Whatever baseline you set in spring 2026 is stale.
The practical takeaway: cover sub-questions (comparisons, pricing, alternatives, how-tos) as separate focused pages, and re-baseline your visibility numbers whenever the engines change behavior. They change behavior a lot.
Step 4: Connect visibility to traffic and revenue
Visibility metrics tell you where you appear. Revenue attribution tells you whether it matters. You need both, and the second lives mostly in GA4.
Build a custom channel group or segment matching AI referrer domains with a regex like:
chatgpt.com|openai.com|perplexity.ai|claude.ai|gemini.google.com|copilot.microsoft.com|grok.com|you.com
Then compare that segment's conversion rate, bounce rate, and average order value against your site average. Add "Landing page" as a secondary dimension and you get a direct map of which funnel-stage content is actually winning AI traffic: blog landings point to awareness, comparison and pricing landings to decision.
Watch for one classic error: AI referral traffic misclassified as "direct" in GA4. If your direct traffic spikes in the same weeks your AI visibility improves, check your channel definitions before celebrating either number.
Is the traffic worth measuring? The benchmarks say yes:
| Source | Conversion rate | Notes |
|---|---|---|
| ChatGPT referrals | 15.9% | Seer Interactive case study, the most-cited primary source |
| Perplexity referrals | 10.5% | 2026 roundups citing Seer's methodology |
| Claude referrals | 5.0% | Same benchmark set |
| Google organic | 1.76-1.8% | Seer Interactive baseline |
The volume is small but the quality is outsized. Ahrefs found 0.5% of AI-referred visitors drove 12.1% of total signups on one site studied. Hat Club found roughly 1 in 50 visitors came from AI referrals, yet that sliver drove 20x revenue growth in AI-driven sales. And the quality has been improving: Adobe Analytics data shows ChatGPT referral conversion went from 43% worse than other channels in July 2024 to positive by mid-2026.
For attribution beyond last-click, tools like Dreamdata or HockeyStack can map AI touchpoints into multi-touch journeys, which matters because AI-influenced buyers often convert through a different channel days later.
What content actually wins citations at each stage
Your content strategy should follow the citation data, and the citation data has shifted hard toward commercial pages in 2026.
Promptwatch's July 2026 data shows product pages were ChatGPT's most-cited content type at 32.8% of citations, nearly double their share since March. In Google AI Overviews, product pages overtook listicles as the single most-cited daily format for the first time in late July (17.9% vs. 16.2% by month-end), after listicles had fallen from ~26% in Q1 to 18%. The old assumption that AI only cites editorial content is dead. Your product and service pages are decision-stage visibility assets, and they need to be measured as such.
Meanwhile, classic consideration-stage formats are growing: within July, listicles grew fastest in ChatGPT (8% to 10%+), how-tos rose from 3.3% to 4.3%, and comparisons from 2.5% to 3.2%. And video is quietly climbing in AI Overviews, from ~2.7% of citations in January to ~6.3% by late July, per Promptwatch's citation-type tracking. Almost nobody is optimizing video for AI citations yet.
On the social side, Reddit still dominates social citations overall (3.36% average share across models), ahead of YouTube at 2.94%, but this varies sharply by engine: ChatGPT is a Reddit specialist (5.19% of citations) while AI Overviews and Grok lead with YouTube. X is essentially never cited, peaking at 0.25% even on Grok. Deprioritize it for AI visibility regardless of stage.
Tools for stage-by-stage measurement
No major platform sells a "funnel stage" tier. Stage segmentation is a tagging feature within prompt sets, so what you're really choosing is tracking depth, engine coverage, and whether the tool stops at monitoring or helps you fix gaps.
A platform like Promptwatch handles the full loop: prompt tracking with volumes and difficulty scores, citation trends classified by content type, competitor share-of-voice heatmaps, and visitor analytics that tie AI visibility to actual traffic and conversions, which is exactly the connection stage-by-stage measurement depends on.
For comparison, here's where the main options stand as of late 2026:
| Tool | Entry price | Engines | Stage-relevant strengths | Weakness |
|---|---|---|---|---|
| Promptwatch | $95/mo | 11+ including AI Mode, Grok, DeepSeek | Crawler logs, visitor analytics, content gap analysis, Reddit/YouTube tracking, agentic content fixes | Newer brand than the SEO incumbents |
| Semrush AI Toolkit | $99/mo | ChatGPT, Google AI, Gemini, Perplexity | Bundles with the full Semrush suite | Fixed prompt sets, shallow AI-specific data |
| Otterly.AI | $29/mo | ChatGPT, AI Overviews, Perplexity, Copilot | Cheapest credible entry point | Gemini, Claude, AI Mode are paid add-ons |
| Peec AI | ~$80-95/mo | 3+ models | Multi-country tracking, Looker Studio export | Monitoring-focused, no crawler logs |
| AthenaHQ | Free tier, then $295/mo | 9 models | Content agent, API access | Thin on decision-stage traffic attribution |
| Profound | $99/mo | ChatGPT (more at enterprise) | Prompt-level visibility score and share of voice | Enterprise positioning, price scales fast |
Otterly.AI
Profound
For the analytics side, pair whatever visibility tool you pick with Google Search Console (still the best free view of AI Overview impressions at the awareness stage) and GA4 for the decision-stage conversion work.
If you want to browse the full category before committing, the AI rank tracking tools directory at ai-rank-tools.com covers the landscape, and bestgeosoftware.com lists GEO platforms with more content-optimization depth.
Common mistakes to avoid
Equal-weighting every prompt. Covered above, but it's the most common and most damaging error. Weight your prompt set toward the stages that produce revenue.
Tracking one engine. Visibility differs sharply by engine, and single-platform tracking hides most of the picture. ChatGPT, AI Overviews, and Perplexity each have distinct citation behaviors, as the Reddit collapse shows: a strategy that worked on ChatGPT in July might still work fine on Google surfaces.
Treating a citation as a win. The 69% citation-without-recommendation figure from Visibility Labs should be on a sticky note on every marketer's monitor. Track recommendation rate separately, especially at the decision stage.
Trusting a single audit. With only 20% of brands remaining visible across five consecutive answer runs, one snapshot tells you about randomness, not visibility. Run a consistent prompt set on a schedule.
Ignoring the awareness blind spot. Referral analytics can't see buyers who read your name in an answer and never click. Survey fields and branded search volume are imperfect proxies, but they're the only ones that catch zero-click awareness influence.
A reporting cadence that works
Here's the rhythm I'd run: weekly prompt-set runs for visibility by stage (awareness mention rate, consideration share of voice, decision first-mention rate), monthly cross-engine trend reviews to smooth out engine volatility, and a monthly GA4 review of the AI referral segment's conversion rate and landing pages. Re-baseline everything whenever an engine ships a behavior change, which in 2026 has meant roughly every couple of months.
The report that comes out of this is three numbers per stage instead of one blended score, and each number points to a different fix. Low awareness mention rate means content and entity problems. High consideration citations with low recommendations means your content informs but doesn't persuade. Weak decision first-mention rate means comparison pages, review presence, and shopping-card optimization. That's a to-do list, not a dashboard, which is the whole point of measuring by stage.