The 5-layer GEO measurement framework for Vancouver SMBs

Most agencies sell AI-visibility services with one number: citation count. “You are mentioned in 12 percent of AI answers about your industry.” The number looks defensible in a slide. For 95 percent of the agencies selling it, that number is not connected to pipeline in any rigorous way. The CFO who learns the difference between presence rate and a closed deal is going to unwind that contract fast.

This is the framework an honest GEO program tracks. Five measurement layers. Each one imperfect on its own. Together they triangulate something real. Adapted for Vancouver SMBs running a marketing budget under $10,000 a month.

Why we need five layers, not one

The honest reason AI search measurement is hard: the platforms do not let you see what they show users. There is no GA4 for ChatGPT. There is no Search Console for Claude. Every measurement is a proxy. Some proxies are stronger than others. None on their own can defend a budget.

So instead of chasing a single closed-loop attribution model, we use five imperfect signals. When they move together, something real is happening. When they diverge, the visibility is vanity.

The goal is not a closed loop because the technology does not allow one. The goal is triangulation: multiple imperfect signals that, when they move together, point to something real.

Layer 1: Direct attribution (GA4 referrers)

Track human clicks from AI tools to your site. By default, Google Analytics 4 bundles about 70 percent of AI-search traffic into the “Direct” channel because the platforms do not always pass clean referrer headers. That number is from Cloudflare’s 2026 traffic data. Two thirds of your AI traffic is invisible until you fix the config.

What to do

• ·Rebuild your channel grouping. In GA4 → Admin → Channel groups → Create a custom group that catches referrers from chatgpt.com, chat.openai.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, and claude.ai. Name the new channel “AI Search”.
• ·Add a custom dimension for full user-agent strings. Helps catch agentic browsers (ChatGPT Atlas reports as Chrome 141) that look like regular Chrome at the HTTP level.
• ·Track conversion events from the AI Search channel. Form submits, phone clicks, booking flows.

Time to implement: one afternoon. Cost: free. This is the cheapest single move with the biggest payoff.

Layer 2: Server log crawl diagnostics

Bot crawls are not the same as user clicks, but they are an early signal of demand. Three buckets matter, and they signal different things.

• ·Training crawlers (GPTBot, ClaudeBot, anthropic-ai, CCBot, Bytespider). Signal: your site is being used to train models. Useful for long-term presence in answer training data.
• ·Indexing crawlers (OAI-SearchBot, Claude-SearchBot, PerplexityBot, DuckAssistBot). Signal: your pages are eligible to be cited when an AI tool searches the web in real time. This is the most directly actionable bucket.
• ·User-triggered fetchers (ChatGPT-User, Claude-User, Perplexity-User, MistralAI-User). Signal: a real user just asked an AI tool a question and the AI fetched your page in real time to answer them. This is gold. Each fetch likely produced a citation in a live answer.

What to do

• ·Pull weekly access logs. Most hosts let you download these. If your site runs on Netlify, the logs live in the dashboard under Analytics.
• ·Parse with an LLM. Paste the log into Claude or ChatGPT and ask it to group by user-agent and URL. Five-minute weekly job.
• ·Verify the bots are real, not spoofers. OpenAI publishes searchbot.json and chatgpt-user.json IP ranges. Anthropic does the same. Cross-reference the IPs in your logs against the published ranges.

Watch for one specific signal: the ratio of crawl volume to referral traffic. Cloudflare data shows Anthropic at 73,000:1, meaning for every visitor Anthropic sends, its bots have read tens of thousands of your pages. Google sits at 14:1. If your site is being crawled heavily by AI bots but you are not seeing referral traffic, the bots are reading you but the models are not citing you. That is a content quality signal.

Layer 3a: Share of voice (across multiple models)

This is what the AI Visibility Score on this site does, and what most agencies call “GEO measurement” even though it is only one layer of five. Ask each AI tool a customer-style question (“Best industry in city?”) and check whether it mentions your business.

Important caveats. Share of voice is polling, not pageviews. Different vendor tools disagree by double-digit percentages on the same brand. Report ranges, not point estimates. And do not call this attribution. It is correlational evidence.

What to do

• ·Run the same question across multiple models monthly, not just one. The AI Visibility Score tool does this across Claude, ChatGPT, Gemini, and Perplexity in 60 seconds.
• ·Track the trend, not the snapshot. A single result is noise. Twelve weeks of monthly snapshots is signal.
• ·Correlate against branded search volume in Google Search Console and direct traffic in GA4. If AI citations rise but branded search stays flat, the visibility is vanity. If both move together, something real is happening.

Layer 3b: AI interrogation (the deeper question)

Presence is one thing. What the AI says about you when asked deeper questions is another. A sales analogy: imagine you sent a brand-new rep to a networking event with no briefing. They will get your positioning wrong, miss the strengths customers actually care about, and quote yesterday’s pricing. You will not hear about it at the event. But you will lose deals from that event for months. AI tools do this at scale, every day, in front of every prospect researching you.

The 5 questions to ask each model monthly

• ·Who is the ideal customer for [your brand]?
• ·What are [your brand]’s strengths and weaknesses?
• ·What problems do [your brand]’s customers face?
• ·Why choose [your brand] over competitors?
• ·What is [your brand] known for?

For each answer, track four things: factual accuracy, ICP alignment, source attribution, and weakness framing. When AI cites outdated information or attributes weak narratives to specific sources, those sources become your content remediation targets.

Layer 4: CRM self-report (the dark funnel)

The most underused measurement layer. Add one form field. The data difference is dramatic.

What to do

• ·Add “How did you find us?” to your contact form. Options: Google search, ChatGPT, Perplexity, Gemini, Claude, Copilot, Bing, friend referral, Other. Plus a short text field: “What did you search or ask?”
• ·Push that data into your CRM as a custom property. Roll it up to deal stage and closed-won value.
• ·Train your SDRs or front-of-funnel staff to ask if the form was skipped. “Quick question, how did you hear about us?” takes 5 seconds and recovers most of the attribution.

Cross-reference self-report against Layer 3a. If branded-search lift and self-reported AI attribution move together, you have triangulation. If they diverge, dig into which one is lying.

Layer 5: Incrementality (the agency’s problem)

This layer is mostly for agencies serving a portfolio of clients. If you run a single business, skip it. The principle is to compare clients with full GEO programs against clients with minimal or no GEO programs, controlling for vertical and starting traffic, and watch the branded-search and pipeline trajectories over 6 to 12 months.

You cannot run a true holdout. You cannot turn ChatGPT off in Cleveland. Control groups are fuzzy. PR, seasonality, product launches all bleed in. Null results are real. A properly run benchmark can show zero measurable lift.

For Vancouver SMBs running their own marketing: do not chase Layer 5 yourself. Watch Layers 1 through 4, and let your agency or fractional CMO own Layer 5 if they claim to.

What the dashboard should look like

Once Layers 1 through 4 are wired up, one screen should show all of it. The visualization that matters most:

• ·AI Search channel sessions and conversions per week (Layer 1)
• ·Indexer vs user-triggered fetcher bot volume on your commercial URLs, with weekly delta (Layer 2)
• ·Share of voice across 4 AI tools, plotted against branded search volume from GSC, with a confidence range (Layer 3a)
• ·AI interrogation accuracy score with source heatmap (Layer 3b)
• ·Percent of closed-won pipeline self-reported as AI-influenced, broken down by tool (Layer 4)

If you cannot build this dashboard yourself, that is fine. The data points themselves are still useful as a monthly report.

Three warnings before you start

• ·Citation share alone is the new domain authority. Looks defensible in a slide, disconnected from pipeline. Do not let an agency sell you GEO based on this number alone.
• ·Fetch volume does not equal traffic. High crawler activity does not guarantee human clicks or even citations. It only proves the bots are interested.
• ·Vendor disagreement is substantial. If you buy a third-party SOV tool, pick one and treat it as a trend instrument, not gospel. Different tools give wildly different numbers for the same brand on the same week.

If you are starting from zero

Do not buy tools. Start with the cheapest layers and graduate up.

• ·Week 1: GA4 channel grouping for AI referrers (Layer 1). One afternoon. Free.
• ·Week 1: Add “How did you find us?” to your contact form (Layer 4). 30 minutes. Free.
• ·Week 2: Run the AI Visibility Score as your monthly Layer 3a benchmark. Free.
• ·Month 2: Start weekly log analysis (Layer 2). Less than an hour to set up.
• ·Month 3: Add Layer 3b interrogation prompts monthly across all 4 models. About 20 minutes a month.
• ·Month 6: Run a 12-week correlation between Layer 3a citations and branded search volume. This is when you can defend a budget.

Bottom line

AI search visibility is real and growing. But you cannot manage it the way you manage Google rankings, because the platforms do not give you the same data. The path forward is triangulation across five imperfect layers. The agencies that build a transparent layered framework now will own credibility when the standards harden. The ones still selling citation count dashboards will get unwound by the first CFO who learns the difference between presence rate and a closed-won deal.

Start with what you can do this afternoon. Channel grouping. Form field. Monthly visibility check. Build the trend line. Six months from now you will have something defensible.