What Does a Generative Engine Optimisation Strategy Look Like?
Nobody publishes how these systems choose sources, so a plan built on assumed mechanics is fragile by construction. The alternative is to order the work by what it costs you if the field changes rather than by what it might return if nothing does.
Strategy Under Uncertainty
An ordinary marketing strategy predicts a return and orders the work by which items promise most. That method needs a known mechanism. In this subject nobody outside the vendors has one, so predicted returns are guesses dressed as forecasts.
What is actually documented. Very little. It is worth naming.
Vendors publish crawler controls and broadly what their products do. None has published how candidate sources are selected or weighted when an answer is composed. Everything else circulating is observed behaviour or invention, as our guide to how these systems assemble an answer sets out.
Why that breaks conventional planning. The inputs are unavailable.
You cannot rank work by expected return when the relationship between the work and the outcome is unpublished. A plan that does it anyway has invented the numbers it is ordering by. It will look confident right up to the point a model update invalidates it.
The alternative. Order by regret rather than by predicted return.
Instead of asking what each item might deliver, ask what it costs you if the assumption behind it turns out wrong. Some work is valuable whether or not any of this develops as expected. Some only pays if a current pattern holds. Some is worthless in every case and a few things are actively harmful.
Why that is the right question here. It needs no mechanism.
Regret can be assessed without knowing how the systems work, because it depends on the alternative uses of the same money and effort. That makes it answerable today, which predicted return is not.
What follows. Four categories.
Blocks two, three and four sort the work into things that help regardless, things that are bets, then things not worth doing. That ordering is the strategy.
This page is reviewed quarterly. The reasoning is stable even where the observations are not.
Work That Helps Regardless
Four workstreams with zero regret attached, because every one of them is justified by ordinary search whatever happens to this field. If the whole subject evaporated next year, none of this would have been wasted.
Clear factual content. Material that answers questions completely.
Written so a passage can stand alone rather than depending on the paragraphs around it. This appears to matter for being used as a source, it certainly matters for ordinary search. It makes the page better for a person reading it. Three justifications, only one of which relies on anything uncertain.
Accurate information about the business everywhere. The same details wherever they appear.
Name, address, services, description, across the site, the profile and third party listings. This mattered for local search long before any of this existed. It appears to matter more when several sources are combined.
Technical accessibility. Including the crawlers belonging to these products.
The only item in this block with documented consequences attached, since vendors publish what blocking their crawlers does. Everything else here is reasoned rather than published.
Genuine third party references. Earned rather than bought.
Our backlinks guides hold the same position they always have, which is that this is a reason to earn references rather than a reason to buy links.
Why this block is most of the plan. Because the regret test says so.
These four carry no downside risk and multiple independent justifications. Any supplier whose proposal skips them to reach the newer material has ordered the work by novelty rather than by consequence.
Work That Is A Bet
These items rest on currently observed behaviour rather than on anything published. They are worth doing. They are worth doing after block two. They are worth reviewing rather than treating as settled.
Structuring pages around questions. A bet with a small downside.
Organising material so headings match real questions with answers immediately beneath. The bet is that extractability matters as much as practitioners currently think. If it turns out to matter less, you are left with clearer pages, which is a modest loss.
Formatting choices. A bet with almost no evidence behind it.
Lists against prose, summary paragraphs, particular heading patterns. Practitioners report preferences here and nobody has published anything. Cheap to do, so worth doing. Not worth building a plan around.
Structured data beyond the basics. A bet requiring care.
No vendor has confirmed structured data is used in generating answers, so we treat it as removing ambiguity rather than as a lever. It also remains an assertion about reality, which means it must never describe something untrue. That constraint is not a bet. It applies regardless.
Optimising for one particular product. The riskiest bet in the block.
Tailoring work to whichever assistant currently matters most assumes that product keeps its position. Given how the landscape has moved, that is a considerable assumption.
How to hold a bet properly. Write down why you did it.
A bet you have recorded as a bet can be reviewed and reversed. One that has quietly become a belief cannot. This field produces those constantly.
Work That Is Not Worth Doing
Three categories where the regret is entirely one sided. These are not marginal calls. Each has a cost and no available upside.
Guaranteed placement, which nobody is able to offer. Not a service.
Nobody controls whether a system names your business, so a guarantee is a claim about something outside the supplier's control. The cost is not only the fee. It is the plan built around an outcome that was never available.
Anything requiring a submission that does not exist. Also not a service.
There is no form, no registration and no console for the products in this cluster. A proposal including submission is describing a step that cannot be performed, which tells you what the rest of the proposal is worth.
Anything that damages the site if the assumption is wrong. The category worth real caution.
Text written for machines rather than readers, instructions hidden in pages, content restructured until it no longer serves a customer. These assume a stable mechanism nobody has published. If the assumption fails you are left with a worse website, which is a certain cost against a speculative gain.
The test that catches all three. One question.
If this field turned out to be less significant than expected, would I regret having done this? For block two the answer is no. For block three it is mildly. For anything here it is yes, which is what puts it in this block.
What to do when a supplier proposes one. Ask what happens if they are wrong.
A capable supplier has thought about it. Our guide to choosing a supplier covers the rest of that conversation.
Deciding What You Are Trying To Achieve
Most businesses spending on this have not decided what they want from it. Three quite different goals get bundled together. They need different work, so picking one changes the plan considerably.
Being cited. Appearing as a source when somebody asks about your subject.
This is a content goal. It needs material that answers questions completely and stands alone. It suits businesses whose expertise is the product. Success looks like appearing in answers about the subject rather than about your business.
Being the recommendation. Being named when somebody asks who to use.
A different goal entirely. It rests on how the business is described across the web, on independent corroboration and on being findable, rather than on how much you have published. Most local and professional service businesses want this one and pursue the first by mistake.
Simply not disappearing. The most common real position.
A business whose traffic came from explanatory pages wants to protect what it has rather than win something new. That is a defensive goal and it points at different work again, mostly at material an answer cannot substitute for.
Why choosing matters more than it sounds. The work diverges.
Publishing more content serves the first and does almost nothing for the second. Fixing the record across the web serves the second and does little for the first. A business that has not chosen buys both badly.
How to choose. By where your enquiries come from.
If people find you by researching a subject, the first. If they find you by asking who is good, the second. If your traffic is falling and nothing else has changed, the third.
Sequencing It
The order matters more than the contents, because several items are useless until an earlier one is done. This is the sequence we would actually run.
First, access. Before anything else, out of order deliberately.
Confirming the relevant crawlers are permitted and arriving. It is documented, it is cheap and everything downstream is wasted if it fails. It also frequently fails for reasons nobody chose.
Second, the baseline. Before making changes.
Recording what can be measured today, however imperfect, so next year's comparison means something. Our measurement guide covers what is actually available.
Third, the record. Wherever the business is described wrongly.
Fourth, the material. Content restructured to answer questions completely.
The largest piece of work for most businesses. The one that overlaps most with ordinary search optimisation.
Fifth, references. Earned over months rather than acquired.
Sixth, the bets. Block three, once the above is done.
Where this sits against ordinary search work. Layered on, not instead of.
For most businesses the fundamentals come first and this is added to them. A plan that treats this as a replacement has misunderstood the overlap, which our comparison of the two maps in detail.
Reviewing It
This field moves faster than anything else we work in, which makes review part of the strategy rather than an administrative afterthought. A plan here has a shelf life.
The cadence. Quarterly for the plan, monthly for the platform positions.
The reasoning holds for quarters at a time. The vendor documentation underneath it does not, which is why the platform guides in this cluster carry monthly review dates while the reasoning pages carry quarterly ones. We tie the plan review to the quarterly technical audit, with written updates every three weeks in between.
What would justify changing direction. Four things.
A vendor publishing something that turns an observation into documented behaviour, which would move an item out of block three. A vendor withdrawing or changing a control, which has happened. A change in which products your customers actually use. Or your own sampling showing a sustained change over months rather than weeks.
What would not justify it. Rather more.
A single month's sampling moving. An article asserting a new ranking factor. A competitor appearing in one answer. A supplier's dashboard showing a different number. None of those is evidence of anything.
The discipline that makes review possible. Recording why.
A plan that says what each item assumes can be revised when an assumption fails. One that lists only tasks cannot, because nobody remembers which parts were bets.
Writing It Down
The document should fit on one page. Anything longer in a field this unsettled is a document nobody will revisit, which defeats the purpose of writing it.
Which of the three goals you chose. One line. The reason.
The work that helps regardless. Listed, with the note that it is justified by ordinary search independently.
This matters when somebody later asks what the AI spend achieved. The answer is that most of it was work you needed anyway.
The bets, each with its assumption written next to it. The most important part of the document.
Not the task. The belief the task depends on. When an assumption changes, you can find every item resting on it.
What you have decided not to do. And why.
Conceding deliberately is what keeps a small budget effective. Writing it down stops the same rejected idea returning every quarter.
What you will measure. Its limits. Stated honourably rather than optimistically.
The review date. An actual date rather than an intention.
Our approach page sets out how we run this for clients, including what we will not claim.
Ordered by regret,
not by prediction.
You cannot rank work by expected return when the relationship between the work and the outcome is unpublished. You can ask what each item costs you if the assumption behind it turns out wrong.
What is included every month:
£350 per month, one target area. No setup fee, nothing billed separately.
Twenty guides.
One subject.
The mechanism, all four platforms, what you can influence, whether it is worth it yet, what it costs, how to measure it and how to judge a supplier selling it.
