Key takeaways
- AI assistants create demand that classic analytics mislabels as 'direct' or 'organic' — the dark funnel is real.
- Measure GEO with a fixed prompt panel: track presence, position, and sentiment across ChatGPT, Claude and Perplexity.
- Perplexity shows its source URLs; use it to learn which of your pages AI models trust.
- Add 'ChatGPT / AI assistant' to your 'How did you hear about us?' field to surface hidden attribution.
- Consistent, factual, well-structured content wins citations — coherence across sources beats clever tricks.
When a prospect asks ChatGPT for the best supplier in your category and your brand gets named, something valuable just happened — but your analytics won't show it. This is the central challenge of Generative Engine Optimisation (GEO): AI assistants like ChatGPT, Claude and Perplexity increasingly shape purchase decisions, yet the citation that drives the sale rarely leaves a clean trail. At Evocative, we treat this as a measurement problem first and a content problem second. You can't optimise what you can't see.
Why ChatGPT citations break traditional attribution
Classic referral tracking assumes a clickable journey: someone clicks a link, lands on your site, and the source is logged. AI search demolishes that assumption. A user might ask ChatGPT for a product recommendation, read your brand name inside the answer, and then — days later — Google that brand directly or type your URL. Your analytics will attribute the visit to "direct" or "organic search," never to the AI conversation that actually created the demand.
This dark-funnel effect is why so many marketing teams feel AI is influential but can't prove it. The recommendation happens inside a closed interface, often with no link at all. And with OpenAI pushing ChatGPT toward a full workspace — Spaces, Pages, autonomous agents connected to thousands of apps — more of the buyer journey is happening inside the assistant rather than on the open web. The click is becoming optional.
A practical GEO measurement framework
You don't need a perfect model. You need a repeatable one that tells you whether your visibility is trending up or down across the major engines.
1. Build a prompt panel, not a keyword list
Instead of tracking keywords, track the real questions buyers ask. Draft 30–50 representative prompts across the funnel — "best X for Y," "X vs competitor," "is X worth it" — and run them regularly in ChatGPT, Claude and Perplexity. Record three things each time:
- Presence: is your brand mentioned at all?
- Position: are you first, in a shortlist, or an afterthought?
- Sentiment & accuracy: is what the model says about you correct and favourable?
2. Track citations and the source behind them
Perplexity is the most transparent — it shows the exact URLs it cites, so you can see which of your pages the model trusts. ChatGPT and Claude are more opaque, but when they do surface links (increasingly via connected tools and browsing), log them. Over time you'll learn which content formats earn citations: structured comparison pages, clear data tables, and genuinely original analysis consistently outperform thin marketing copy.
3. Instrument the downstream signal
Since the click often arrives later and elsewhere, watch for proxy signals: a rise in branded search volume, direct traffic spikes that correlate with AI visibility gains, and self-reported attribution. The single most underrated tool here is a "How did you hear about us?" field on your forms — add "ChatGPT / AI assistant" as an explicit option and the dark funnel starts to light up.
From measurement to influence
Once you can see your citation share, the optimisation levers become obvious. Models cite sources they can parse and trust, so publish clear, factual, well-structured content with defensible claims. Keep your core facts — pricing logic, service scope, differentiators — consistent across your site, third-party listings and review platforms, because LLMs synthesise from many sources and reward coherence. And monitor for errors: an AI confidently stating something wrong about your brand is a reputational issue you can only fix if you're measuring in the first place.
The engines themselves are moving fast. OpenAI's recent push toward always-on agents and an in-chat office suite means ChatGPT is no longer just an answer box — it's becoming a workspace where decisions get made. The brands that win won't be the ones that gamed a ranking; they'll be the ones consistently cited as the trustworthy answer, measured engine by engine, prompt by prompt.
Where to start this week
Pick ten high-intent prompts, run them across all three assistants, and log your presence and position in a simple spreadsheet. Do it again in 30 days. That delta — your citation trend — is the first real GEO KPI most teams have ever had. Everything else builds from there.
Sources
- Le résumé des annonces d’OpenAI lors du DevDay 2026 : dots, ChatGPT Space, Codex… — Blog du Modérateur
- OpenAI lance les dots, des agents IA autonomes connectés à plus de 4 000 applications — Blog du Modérateur
- ChatGPT Space, Pages et slides : OpenAI dessine sa propre suite bureautique — Blog du Modérateur
- Ce qu’on a lu, vu, aimé et entendu en septembre 2026 : les recos de la rédaction — Blog du Modérateur
- IA : les 10 générateurs d’images les plus performants en septembre 2026 — Blog du Modérateur
- What’s Making AI Impact Difficult to Measure for Marketers? — Neil Patel
