SEO Has a Measurement Crisis. Here’s What I Mean, What To Change, and the Slides

Please note, for speed, I have used AI to generate this post based on research I have performed for my brightonSEO April 2026 talk.

If you saw my brightonSEO April 2026 talk, this post is the follow-up.

If you didn’t, the short version is this:

SEO has not become less effective. It has become less observable. That is the core argument of the talk, and the reason so many SEO teams now find reporting harder, stakeholder conversations harder, and simple “rankings up = traffic up = value up” explanations less convincing than they used to be.

LinkedIn post that inspired the talk: “I need to admit something about SEO” LinkedIn post

You can get the slides here: SEO Has a Measurement Crisis brightonSEO April 2026 slides

How to Track the ROI of SEO Without Pretending the Click Tells the Whole Story

The problem is not that SEO stopped working

The problem is that a growing share of SEO influence now happens in ways that do not show up neatly in classic click-based reporting.

That is the thread running through the talk.

Search results have become discovery surfaces. Users can see brands, compare options, form impressions, and even get answers before they ever click through to a site. In that world, a lot of reporting still starts too late. It starts at clicks, sessions, conversions, and revenue, and then acts surprised when the explanation feels thin.

That is why SEO teams increasingly see the same pattern:

rankings improve
impressions improve
clicks do not follow in the old way

And that is why the internal conversation becomes painful:

“Why are rankings up but traffic flat?”
“Why should we invest if clicks are down?”
“Is SEO actually still working?”

Modern SEO is more probabilistic than deterministic

Old SEO was much easier to explain because the model was simpler.

Rank.
Click.
Session.
Conversion.

That model was never perfect, but it was workable because search itself was simpler and a click was a much better proxy for attention than it is now. Modern search does not behave that neatly. Users compare, browse, remember, return later, and convert through messier, delayed paths. That is why modern SEO is more probabilistic: rankings and visibility still matter, but they now increase the likelihood of influence rather than guaranteeing one neat, proportional outcome.

That difference matters because it changes what reporting can reasonably prove.

The evidence is real, but still early

One of the points I wanted to make in the talk is that this is not just theory anymore. There are examples linking increased AI visibility or citation share to leads, pipeline, and revenue. But the evidence base is still small, often agency-led, and not strong enough to pretend we have perfect causality.

A few examples from the research:

In one B2B SaaS case study, citation rate increased from 8% to 24% in 90 days, producing 47 AI-referred leads, more than €180K in projected pipeline, and €64K in closed sales from a €16K spend. The write-up describes this as 288% ROI, though it is still an agency case study rather than an independent experiment.
In a retail-tech SaaS case, one agency reported 13x LLM-sourced revenue year over year after improving AI answer visibility, again with the caveat that attribution remains uncertain and the figures are self-reported.
In an ecommerce case, AI-driven visits reportedly increased 693%, with AI traffic converting at 5% vs 4% for organic search and AI-driven revenue increasing 120% in four months.

So the right conclusion is not “this is fully proven.”

It is:

there is now enough evidence to take the measurement problem seriously, but not enough to fake certainty.

The more useful model: Presence, Preference, Performance

The framework I use in the talk is simple:

Presence

Did we show up?

This is the visibility layer:

rankings
AI answers
SERP features
impression trends

Preference

Did people lean toward us?

This is where visibility starts to become behaviour:

brand search
direct return visits
assisted conversions
repeat visits

Performance

Did value follow?

This is the familiar layer:

clicks
sessions
conversions
revenue

The reason this lens matters is that most dashboards still start at Performance. That means they often miss the earlier layers where influence begins. And when those earlier layers are invisible in the report, SEO can look weaker than it is.

What to change next week

The talk ends with a practical recommendation.

If you want to improve reporting quickly, start with three things:

1. Separate reporting by layer

Do not force rankings, visibility, traffic, and conversions into one messy story.

Split them into:

Presence
Preference
Performance

That alone makes reporting more honest and easier to explain.

2. Add preference signals into reporting

Most teams can do this now.

Start with:

branded search
direct return visits
assisted conversions
repeat visits
self-reported discovery if you can collect it

These are not perfect proof. They are clues that visibility may be starting to change behaviour.

3. Stop treating traffic as the whole story

Traffic still matters. But it is no longer the full scoreboard.

In answer-led search environments, you can gain visibility and still lose clicks. The better question is not just “did traffic go up?” It is:

where did we show up?
what behaviour changed?
what outcomes followed?

What to build next quarter

The next step is not a giant rebuild. It is a simple influence view.

That means one reporting view that connects three questions:

Where did we show up?

This could include:

rankings
impressions
SERP features
AI mentions
citations
local visibility
product visibility

What matters will vary by business model.

What behaviour changed?

This is the middle layer:

branded search moved
direct visits increased
users returned
assisted conversions rose
people came back later through another channel

This is the layer where you start to see whether visibility is shaping preference.

What outcomes followed?

This is the commercial layer:

leads
demo requests
calls
bookings
sales
revenue
pipeline

This is still the layer the business cares about most. It just needs more context around it than it used to.

Direct AI traffic is real. It is also incomplete.

One of the more useful ideas in the research is the distinction between:

directly measurable AI referrals
AI-influenced sessions that are not directly attributable

The first group is the easy bit: sessions where a referrer is preserved and analytics can see traffic coming from an AI product. The second group is what some people are starting to call dark AI: journeys where the user saw the brand in an AI answer, did not click, and later came back via branded search, direct, or another channel.

That matters because GA4 can report the first group reasonably well, but not the second.

The practical recommendation from the research is:

create a dedicated AI assistants channel in GA4 using custom channel groups
report sessions, engagement, conversions, and revenue for that channel
treat it as a lower bound of AI-driven discovery, not a complete accounting of AI influence

That is a much more credible way to talk about “AI traffic” than either ignoring it or pretending it explains everything.

The most useful measurement stack right now

If I were building this from scratch today, I would use a simple stack:

1. GA4 for directly attributable AI referrals

Track:

sessions
engaged sessions
conversions
revenue
landing pages

Do this with a custom AI assistants channel and update the rule set over time.

2. Search Console and analytics for preference signals

Track:

branded search impressions and clicks
direct return behaviour
assisted conversions
repeat visits

Overlay those trends with AI visibility work and major content changes.

3. AI visibility tracking for inclusion, not just clicks

Use prompt sampling or specialist tools to track:

mention share
citation share
prompt-level share of voice
answer volatility over time

The important caveat here is that prompt share of voice is not traffic. It is inclusion probability and competitive shelf space.

4. Build a small triangulation model

Do not ask one metric to do everything.

Look for patterns:

visibility rises
brand demand rises
conversions or pipeline rise

That is not courtroom evidence. It is business evidence. And in a messy, zero-click environment, that is often the more useful standard.

You do not need perfect attribution

This is probably the most important point from the talk.

You do not need perfect attribution to make better decisions. You need enough evidence to interpret what is happening honestly. The talk calls this directional truth: a better read on reality, enough clarity to explain performance, and enough confidence to keep moving.

That is a better standard than fake certainty.

It is also much more useful in a world where:

no-click behaviour is normal
AI referrals are measurable but still small
visibility increasingly happens inside answers
and the old “one metric tells the whole story” habit has clearly broken down.

Final thought

The simplest way I can put it is the final line from the talk:

Treat SEO like a system, not a vending machine.

For a long time, SEO was treated like a neat input-output machine:
put rankings in, get conversions out.

That is not how it behaves anymore.

Now it behaves more like a system:

influence is delayed
value shows up in different places
the evidence is messier
and explanation matters more than it used to

But the work is still real. The value is still real. The measurement model just needs to catch up.

Links

Slides from the talk:
SEO Has a Measurement Crisis brightonSEO April 2026 slides