Your traffic is falling and your pipeline is fine. That is the story

I have had four versions of the same call this year. Organic sessions are down twenty to forty percent year on year. Rankings are broadly stable. Pipeline is flat or slightly up. Somebody wants to know what broke.

Nothing broke. The top of the funnel moved into a place where sessions are not generated. A buyer who would once have clicked four blue links now asks an assistant, gets a synthesised answer with three brands named in it, and only visits a site when they have already decided who is worth visiting. That shift shows up in analytics as a decline and shows up in reality as a different kind of research, not less of it.

Which pages lost the chatgpt traffic to your website

The decline is not evenly spread, and the shape of it is the diagnosis. Split your organic pages into two groups and look again before you assume the worst.

The genuinely worrying pattern is the inverse: commercial pages declining while informational pages hold. That usually means you are being left out of the shortlist an assistant produces, and the buyer is arriving at a competitor's comparison page instead of yours.

Losing traffic that never converted is not a crisis. Losing the shortlist is, and it does not show up in a sessions chart.

The measurement that replaces sessions

1. 1

   Prompt coverage

   Write the twenty questions a real buyer asks in the ninety days before they choose a vendor in your category. Run them across the major assistants on a fixed schedule and record whether you are named, what you are named for, and which source the model leaned on. This is the closest available equivalent to a rank tracker.

2. 2

   Branded demand

   Branded search volume, direct traffic and the share of demo requests that arrive already naming you. When answer engines do their job, they produce name recall rather than clicks, and this is where it lands.

3. 3

   Self reported attribution

   One free text field on the demo form. It is imprecise, unmodellable and by far the most honest data you will collect. When buyers start writing 'ChatGPT suggested you' in that box, you have your answer.

What I would not do

• Do not publish more definitional content to win the traffic back. That is exactly the content the assistant absorbed, and republishing it feeds the thing that replaced you.
• Do not block the AI crawlers in retaliation. You cannot be cited by a system that is not allowed to read you, and the traffic does not come back either way.
• Do not set a target to restore last year's session count. You will hit it with traffic that costs money and produces nothing.

Why ai search referral traffic looks tiny even when it matters

If you go looking for ai search referral traffic in your analytics right now, expect a small number. Most platforms answer inside their own interface and only pass a referrer to your site when a user actively clicks a citation link, and most users do not click, because the answer already satisfied them. A small referrer count does not mean small influence. It means the influence happened somewhere your tools cannot see.

This is the same measurement gap that made brand advertising hard to prove for decades. Nobody could show the board a session for a billboard either, so the industry built proxies: aided recall, branded search lift, share of voice. AI visibility needs the same kind of proxy thinking, not a request for a bigger referrer number that is never going to show up.

The diagram of where the buyer actually is

Look at that sequence and the sessions problem makes more sense. Your analytics only sees step four and step five. Steps one through three, where the shortlist actually gets decided, are invisible to a platform that only counts page loads. If you are not in the shortlist at step three, you never get a session at step four to lose, and no session decline will tell you that. This is why prompt coverage matters more than any traffic report: it is the only place you can see step three happening at all.

The counter-argument worth taking seriously

The uncomfortable part of this shift is that it removes a number that made marketing legible to people outside marketing. Sessions were easy to explain and easy to trend, one line on a chart going up or down. What replaces them is messier: a coverage score across a rotating set of prompts, a branded demand line that moves slowly, and a stack of verbatim quotes from buyers that has to be read rather than graphed.

Messier and closer to the truth is a good trade. It was always a proxy. The job was never traffic, it was being the name that comes up when someone with a budget asks who solves this.

Why the click was never the real signal

Go back further than this year and the click was always a workaround. Buyers wanted an answer. Search engines could only hand over a list of documents, so the click was the tax you paid to get from the list to the answer. Assistants remove the tax for a large share of questions. That is a better outcome for the buyer and a worse one for anyone whose reporting depended on counting the toll booth.

The categories most exposed are the ones where the answer is short and stable: definitions, pricing ranges, feature comparisons that do not change monthly. The categories least exposed are the ones where the buyer needs to see something for themselves, a live demo, a specific integration, a number that applies to their exact situation. Map your content against that line and you will find most of the traffic anxiety sits in the first group.

How to set up the tracking this week

llm traffic analytics is still an underbuilt category. Most analytics platforms bolt on a referrer filter for the big assistant domains and call it done, which misses the sessions that never had a referrer to begin with. Until better tooling exists, the spreadsheet and the weekly prompt run are not a stopgap, they are the state of the art, and teams waiting for a vendor to solve this are waiting for something that will arrive after the gap has already cost them a shortlist slot.

What to tell the board this quarter

If someone senior asks why organic traffic is down, the honest answer has three parts. First, the decline is concentrated in pages that were not producing pipeline. Second, we have a new tracking method, prompt coverage, that shows whether buyers still find us before they ever land on the site. Third, we are reallocating the hours that used to go into chasing session volume toward the pages and off site sources that actually influence the shortlist. That is a shorter conversation than defending a chart that is trending the wrong way for reasons nobody in the room can see.

Where seo beyond google actually starts

Thinking about seo beyond google is not a separate discipline from search optimisation, it is the same discipline with a wider set of engines reading your content. Google still crawls, indexes and ranks the same pages. What changed is that three or four more systems now also read those pages, decide whether to trust them, and skip the ranking step entirely by answering the buyer directly. Your technical foundation does not need to change. Your assumption about what happens after the crawl does.

Practically, that means the audit checklist grows by one column. You still check crawlability, indexation and page speed. You add a check for whether the page states a complete, verifiable answer in its first two sentences, because that is the unit an assistant lifts, not the page as a whole.

What this means for the content calendar

Most content calendars are still built around keyword volume, which is the wrong input now for a meaningful share of topics. A term with 2,000 monthly searches that resolves in a single sentence is not worth a new article, because the article will not be read, it will be summarised and discarded. The volume number tells you demand exists. It does not tell you whether that demand still routes through a page.

The better filter is whether the question needs proof. Does answering it require a number specific to the reader's situation, a comparison they need to see laid out, or a demonstration they need to watch. If yes, write the page, because that is exactly the content an assistant cannot fully substitute and will instead cite. If no, the honest answer is to stop writing it and put the hours into the pages that do the citing job.

• Kill or consolidate thin definitional pages that exist only to catch long tail volume.
• Invest the freed hours in comparison, pricing and methodology pages that a buyer has to open to verify.
• Treat any new definitional content as a citation play, not a traffic play. Write it to be quoted, not to be visited.