How do we tell whether it is working?

Not from visits, which is the whole difficulty. Ask the systems the questions your buyers ask and read what comes back, watch whether people arrive already using your framing, and accept that this is softer evidence than a click was and better than a number that cannot distinguish success from failure.

What techniques actually help?

The ones indistinguishable from doing the work well: clear statements, material that is correct, and being the place a specific thing is explained properly. Anything presented as a mechanism separate from that deserves the question of what evidence supports it.

Is being cited without a visit worth anything?

It is worth something and not the same thing. Shaping what people are told is genuine influence and it does not pay a bill directly, so a business that needs the visit is facing a positioning question rather than an optimisation one.

Glossary

Generative engine optimisation (GEO)

Making content likely to be summarised and cited by generative systems, where being the source of an answer matters more than being visited.

In plain terms

Trying to be the source a generated answer draws on. The awkward part is that succeeding produces no visit, so the thing everybody has always used to tell whether their content worked simply does not happen.

Why it matters

Because the measurement problem is the real change and it gets much less attention than the techniques. An organisation can be doing better than ever at influencing what people are told and see a falling line on every chart it has.

How it works

Success here does not produce a click. If your page is the source a summary drew on and the reader was satisfied, they never arrived, so the event your entire measurement rests on did not occur.

That makes the instrument the problem rather than the technique. A team judging content on visits is now measuring a shrinking part of its own effect, and no adjustment to how the content is written changes what the chart is counting.

What can be observed instead is indirect and worth setting up anyway. Whether people arrive already knowing your framing, whether your terminology appears in their questions, and whether being cited correlates with anything downstream are all softer and all real.

Asking the systems is a crude check that costs nothing. Putting the questions your buyers ask into a few of these tools and reading what comes back tells you whether you are represented and how, which is the closest thing to a rank check available.

The techniques, where any are defensible, collapse into being genuinely worth citing. Clear statements, material that is actually correct, and being the place where a specific thing is explained properly are what survive summarisation, and none of that is separable from doing the work well.

The lag is longer than anybody is used to. Material has to exist, be found and be drawn on before any of this is observable, and none of those steps happens on the schedule a campaign is reported against, so a monthly review is measuring noise.

Being cited and being visited are different goods, and it is worth deciding which you are pursuing. A reputation for being right is worth having and does not pay a bill directly, so a business that needs the visit has a positioning problem rather than an optimisation one.

The same chart, two opposite meanings

The temptation, faced with an ambiguous measurement, is to find a better one, and there is not currently a good one to find. What there is instead is a set of weaker observations that together lean one way or the other, and the honest recommendation is to gather two or three of them rather than to keep staring at the traffic. Ask the tools your buyers' questions once a month and keep the answers, so you can see whether your framing appears and whether it is accurate. Ask the people who do arrive what they already knew, which takes one extra question in a call somebody is having anyway. Watch whether your own terminology comes back to you in the questions people ask. None of that is a metric anybody would defend in isolation, and the combination is considerably better than a number that cannot distinguish success from failure. It is also, unglamorously, how most people judged whether their writing was working before anybody could count clicks.

Seen in the wild

Asking a generative tool the questions your buyers ask, and reading what it says about you.
Perplexity
A definition page cited in answers, generating no visit and shaping the framing.
ChatGPT
Terminology from your material appearing in the questions people bring you.
Writer

Common misconceptions

People assume

It is a set of techniques like the old ones.

In fact

The techniques that survive scrutiny reduce to being worth citing. The genuinely new problem is that success is invisible to your existing measurement, and no technique addresses that.

People assume

Falling visits mean the content is failing.

In fact

It can equally mean the content is working and being consumed elsewhere. Those are opposite conclusions from the same chart, and distinguishing them needs a different observation rather than a closer look at the same one.

Questions

How do we tell whether it is working?: Not from visits, which is the whole difficulty. Ask the systems the questions your buyers ask and read what comes back, watch whether people arrive already using your framing, and accept that this is softer evidence than a click was and better than a number that cannot distinguish success from failure.
What techniques actually help?: The ones indistinguishable from doing the work well: clear statements, material that is correct, and being the place a specific thing is explained properly. Anything presented as a mechanism separate from that deserves the question of what evidence supports it.
Is being cited without a visit worth anything?: It is worth something and not the same thing. Shaping what people are told is genuine influence and it does not pay a bill directly, so a business that needs the visit is facing a positioning question rather than an optimisation one.