Two kinds of conversation have been landing with us this year. Some businesses arrive having watched Search Console tell them the same story for months - impressions holding up, clicks drifting down, and nothing wrong with the rankings to explain it. Others arrive from the opposite direction: they already run Google Ads with us, they've noticed assistants answering their buyers' questions, and they want to be in those answers before their competitors are.
Both end up in the same place, which is why it's worth separating two things that usually get argued as one. Some traffic genuinely has gone. Rather more is still winnable than the average LinkedIn post about this suggests.
The traffic that isn't coming back
Someone searched for a quick fact, got it in the Overview, closed the tab. That visit is gone, no amount of optimisation retrieves it, and anyone telling you otherwise is selling something.
It lands hardest on one type of content: the high-volume informational post that existed to catch a definition query and funnel some percentage of readers somewhere more useful. Whole content programmes were built on that mechanic. A fair few of them no longer work, and the sensible response is to stop resourcing them rather than to publish more of them, louder.
Citations are the contested ground now
The answer has to come from somewhere. Overviews cite sources, assistants cite sources, and those slots are being claimed while a lot of businesses are still deciding whether any of this is a fad.
Being the cited source is worth two fairly different things. There are the clicks that still happen, from people whose question wasn't fully resolved and who want to see where it came from - fewer than before, though usually further along than an average organic visit. And there's presence inside the answer itself, where your name sits next to the recommendation whether or not anybody clicks.
The second is harder to value and probably matters more. It's the same mechanism as being the company ChatGPT names when a buyer asks who to shortlist: one answer, a few sources, and either you're in it or a competitor is.
What earns the slot isn't exotic. Crawlable, well-structured pages; consistent entity signals so the systems can tell what you are; content that makes clear claims and answers questions directly enough to be lifted without mangling; third-party corroboration backing up what you say about yourself. Most of that overlaps with SEO and with the structured data work we've been doing for years, which is why we run it as one programme rather than sell it as a separate product. None of it guarantees placement, because nothing does.
Measurement has to widen
If clicks are a narrowing window onto your visibility, measuring clicks alone will show you a decline that's part real and part artefact of the format, without distinguishing between them. Ours has three parts.
Structured prompt testing across the major assistants: whether you come up for your buyers' actual questions, how you get described when you do, and who gets named instead when you don't. Tracked over time, because any single run is noise.
AI referral traffic in analytics, separated out and watched for what it does after it lands. The volumes are usually small enough that people wave them away. The behaviour is usually better than average, which is the more interesting number of the two.
And, where we have Search Console access, reading it for the new pattern rather than the old one. Impressions holding steady while clicks thin out is a format change, not a ranking problem, and treating it as a ranking problem sends you off fixing things that were never broken.
This is early-days measurement and we say so. I'd rather hand over limited numbers with the caveats attached than a confident dashboard implying more precision than actually exists.
What this changes in the reporting
Reporting built purely on sessions will keep showing a decline it can't explain. Adding citation presence to the monthly report is a small change that makes the rest of the numbers readable again.
Most clients aren't doing it yet. Starting now mostly buys you a baseline, which is unglamorous, but there's no showing a trend without one.
