I've been involved in SEO for well over a decade, and for most of that time, measuring search visibility was challenging but legible. Monday morning meant opening Google Search Console, checking impressions, confirming that position 4 was still position 4. Your colleague in Manchester ran the same query and saw the same result. The feedback loop was slow, but you trusted it.
Then someone asked you about AI search visibility. Maybe it was a quarterly review, maybe it was an offhand question in a Slack thread. And you realised you didn't have a clean answer. Not because you hadn't been paying attention, but because the measurement infrastructure you'd spent years building simply doesn't cover the new surface.
What Made the Last Decade of SEO Stable
Traditional search is deterministic. Google's algorithm applies consistent ranking logic to a fixed index and returns results in a predictable order. Within a market, position 1 is position 1. Your competitors see the same landscape you do.
This made measurement tractable. GSC gave you impressions, clicks, and average position going back sixteen months. Semrush and Ahrefs showed you competitors' estimated traffic and keyword rankings. You could build a content calendar around gaps in the data. You could run a test: publish a piece, wait eight weeks, check whether the target query moved, and eventually read a result.
Marketing teams built entire operating rhythms around this. Weekly rank checks. Monthly reporting decks. Quarterly content audits. The SEO discipline had a stable scaffold, and agencies and in-house teams alike knew how to work within it.
GEO and AEO Are Familiar in the Right Ways
If you've built a serious SEO content strategy, the transition to GEO and AEO isn't as foreign as it might feel. E-E-A-T signals (named authors, cited sources, institutional credibility, specific claims backed by data) matter just as much for AI citation as for organic rankings. Technical hygiene carries over directly. A well-structured long-form piece with clear section headers and direct answers is well-positioned for both featured snippets and AI citation.
The craft of writing to be found is the same craft. What doesn't carry over is the measurement model.
The Signal Is Broken in a New Way
AI responses are personalised. Two people asking ChatGPT the same question get different answers, shaped by their conversation history, account settings, the time of day, and the non-deterministic sampling built into every large language model. There is no canonical AI response the way there is a canonical Google result.
Citation is binary and contextual. Either your brand was mentioned or it wasn't, and whether it appears next time is genuinely uncertain. There's no position 3 in AI search, no impressions metric, no click-through rate. Perplexity doesn't email you a weekly report.
The feedback loop on content changes is worse. With traditional SEO, you knew what you were waiting for: Google's crawler to re-index the page, the algorithm to reconsider its ranking. With AI search, retrieval pipelines update silently and on unpredictable schedules. ChatGPT's web search layer, Claude's Brave Search integration, and Google's AI Overviews each operate on different logic, with different freshness characteristics, and none of them tell you when or why your citation status changed.
You're No Longer Optimising for One Thing
For the last decade, "search" meant Google. Bing existed, but for most brands, winning on Google was the game. One algorithm, one toolset, one set of guidelines.
AI search has no equivalent centre of gravity. ChatGPT accounts for roughly 87% of AI referral traffic right now, but Perplexity is growing fast, Claude has a meaningfully different citation pattern, and Google AI Overviews sits inside the search experience billions of people already use. These are not the same product with minor differences.
Research from the University of Toronto found that Claude shows stronger cross-language domain stability than any other AI engine, consistently returning to the same authoritative sources across languages. ChatGPT switches site ecosystems by language more than any other platform. Perplexity surfaces citations prominently by design and tends to refresh faster than the others.
That has practical implications for where to put effort first. A rough heuristic: if you're a B2B brand with an international audience and a long consideration cycle, Claude's consistency makes it worth prioritising in your GEO monitoring. If you're a consumer brand optimising for volume, ChatGPT's referral dominance makes it the primary surface to track. If your content is research-heavy or technical, Perplexity's citation-first design and faster refresh rate means it's more responsive to content changes.
Being cited on one tells you almost nothing about your standing on another. The monitoring surface has multiplied, but most marketing teams are still running a single-platform operational model.
The Chaos Is Structural, Not a Maturity Problem
It's tempting to think this is a measurement gap that purpose-built tooling will eventually close. That once the category matures, there'll be a clean analytics layer for AI visibility the way GSC exists for organic. Some of that will come. But it won't restore the determinism of traditional SEO, because that determinism was a property of the underlying system, not the tooling.
LLMs are probabilistic by design. Retrieval-augmented generation adds another layer: which documents get surfaced depends on embedding similarity, recency, and relevance scores that shift as models update. Context windows, system prompts, and conversational memory all interact in ways publishers don't control and can't observe.
Consider a mid-size B2B SaaS with solid organic rankings and healthy GSC numbers. On paper, search looks fine. But a competitor has started appearing by name in ChatGPT responses to their core category queries. The team has no way to know this is happening, how often, or what content change triggered it. Their existing stack, built for a deterministic system, has no surface for this. That gap isn't a tooling failure. It's structural.
Building the Stack
You need more measurement points, not fewer. A single query, on a single platform, on a single day tells you almost nothing. What you want is a pattern: citation rate across a representative query set, across multiple platforms, tracked consistently over time.
You also need to separate what you can control from what you can only observe. You can control content quality, crawlability, schema markup, earned media coverage. You can't control whether ChatGPT cites you in a given response. Invest in the controllable inputs, measure the observable outputs, and accept that the connection between them is probabilistic.
The stack that covers this has four layers:
- Google Search Console for organic performance and AI Overview eligibility
- A rank tracker (Semrush, Ahrefs, or equivalent) for AEO surfaces: featured snippets, PAA boxes, position tracking
- Your analytics platform to confirm whether AI referral traffic is actually converting
- A GEO-specific tool for cross-platform citation rate, which is the surface none of the first three were built for. That's the gap Citation Hawk fills: running your tracked queries weekly across ChatGPT, Claude, and Google AI Overviews and showing you citation trends over time. It sits alongside the other instruments, not above them.
The legibility of the last decade isn't coming back. On Monday morning, that means opening four tabs instead of one, reading probabilistic signals instead of clean rankings, and reporting trends instead of positions. It's more work. But the alternative is flying blind on a surface that's already driving meaningful traffic for the brands paying attention to it.
Tom Eastwood is the founder of Citation Hawk, a tool that monitors AI citation rates across ChatGPT, Claude, and Google AI Overviews.