How Lillian Purge Approaches Generative Engine Optimisation
What we do, what we will not claim and what we cannot measure. In a field this full of overclaiming, restraint is the only differentiator worth having. Every commitment on this page is true of every contract we issue.
Our Position
Most of this work is rigorous search optimisation, plus consistent information about your business, plus being referenced by other people. The novelty is in the outcome rather than in the method. We say that before quoting rather than after being asked.
What genuinely differs. Three things, no more.
The outcome is a citation rather than a click. Corroboration across independent sources appears to matter more, which is observed rather than documented. And the measurement problem is unsolved. Our pillar guide sets out the full argument.
What follows commercially. We do not charge twice.
Where an activity is the same activity under two names, it is billed once. We will not present content quality, accessibility or citation work as new because it is described under a newer heading.
Why we take this position rather than the profitable one. It is checkable.
Google's own Search Central documentation, checked on 30 July 2026, states there are no additional requirements to appear in its AI features and no other special optimisations necessary. A firm claiming otherwise is contradicting the largest vendor in the field, in writing, where anybody can look.
What this page is. Commitments rather than description.
Everything below is true of every contract we issue. If a line could not be honoured on every job it is not on the page, which is the same standard we apply to our backlinks and agency pages.
This page is reviewed quarterly. With the rest of the reasoning pages.
What We Do Before Proposing Anything
Three checks, in this order. They are quick, they cost you nothing and they frequently end with us telling you the money belongs somewhere else first.
One. Are the fundamentals sound? Accessible, consistent, answering something.
Can your pages be reached and found. Does your business information agree with itself wherever it appears. Does your content answer any question a customer actually asks. If any of those fails, that work comes first and this subject waits.
Two. Is your category actually affected? Checkable in an afternoon.
We ask the questions your customers ask, in the products they use, then see whether anything substantive appears. Google's documentation, checked on 30 July 2026, states its AI Overviews often do not trigger, so this is a real question rather than a formality.
Three. Can you attribute an enquiry? The one people overlook.
If you cannot currently tell where enquiries come from, nothing here can be judged. Fixing that is cheaper than anything else we could sell you and it comes first.
What we do when a check fails. Say so.
We tell you which one failed and what to do about it, whether or not that leads to us being paid. Our page on whether this is worth it names the groups we would tell to wait.
How often this happens. Frequently enough to be the normal outcome.
What The Work Actually Involves
Five workstreams, with the overlap stated rather than hidden. Three are additional to ordinary search optimisation. Two are not additional at all.
Crawler access, checked and maintained. Additional.
Permitted at every layer, confirmed arriving in your server logs, re-checked after any hosting or security change. The only part of this subject with documented consequences attached.
Content restructured for standalone passages. Partly additional.
Written so a passage answers a question without depending on the paragraphs above it. Additional in requirement, shared in benefit, since it also makes the page better for a person reading it.
Observation and reporting. Additional.
Sampling, server logs and a recorded baseline, because conventional reporting covers none of it.
Entity consistency. Not additional.
The same business details wherever they appear. This was local search work long before any of this existed.
Earned references. Not additional.
Independent sources describing you, earned rather than bought, on exactly the position our backlinks guides have always held.
Why we publish the split. So you can check the invoice against it.
A supplier who will not tell you which parts are additional has removed your ability to judge the price. Publishing it means any client can ask why an item appears under this heading rather than under the search work they already pay for, which is a question we would rather answer in advance than be asked in month three.
What We Will Not Claim
Each of these is a commitment rather than a preference. Each has a reason attached. A refusal without a reason is a marketing position.
We will not guarantee an appearance in any assistant. Because nobody controls it.
No vendor has published how sources are selected when an answer is composed. None offers a mechanism for inclusion either. A guarantee would therefore be a claim about something outside our control.
We will not present a proprietary score as a measurement. Because we would have to invent the method.
Everything measurable here could be combined into an index and given a name. The number would look like a measurement while being an opinion with arithmetic attached, so we do not produce one.
We will not state how these systems work internally. Because it is not published.
Where we describe behaviour, it is either attributed to a vendor with a date or labelled as observation. Where we do not know, we say so rather than filling the gap.
We will not promise a timescale we cannot support. Because there is no dataset to derive one from.
We commit to the order in which work happens and evidence appears, which is knowable. Our page on timescales sets out that sequence and what would show it failing.
And one more, which is not about claims. We will not name a tool as though the tool were the service.
How We Measure It
Five inputs, each reported with what it cannot establish. There is no position to report and no citation count, so what follows is proxies and samples described as such.
Crawler access from your server logs. Hard evidence, limited to access.
Which crawlers reached your site, which pages and what your server returned. This proves eligibility to be retrieved. It does not prove you were used.
A fixed question set, sampled monthly. With the questions shown to you.
You see the set, so you can judge whether they are the questions your customers ask. It is a sample of a variable output rather than a measurement.
Search Console performance. From the console you own.
Google's documentation, checked on 30 July 2026, states sites appearing in its AI features are included in overall search traffic there. Those clicks are not separated from ordinary ones, which is a real limitation.
Branded and non branded search, separated. Always reported apart.
Searches for your name grow for reasons unconnected to this work, so mixing them would flatter us.
Enquiries, counted. The figure that matters, weakest in attribution.
Nobody can prove a given enquiry came from being cited. Our measurement guide sets all of this out in full.
How It Is Budgeted
Inside the monthly retainer, at £350 a month for one target area, with no setup fee and nothing billed separately. There is no premium line for this and there will not be one.
Why we price it this way. Because of the overlap.
Three of the five workstreams in block three are additional and two are not. Billing the whole thing as a distinct service would mean charging separately for work you are already paying for.
What a separate line would incentivise. Worth naming.
A supplier billing this separately has a commercial interest in the work appearing as different as possible from ordinary search optimisation. That incentive points away from accuracy, which is a reason to avoid the arrangement rather than to police it.
The weakness of our approach. Stated because it is real.
Including it means you cannot see what proportion of a month went to this rather than to something else. That is a genuine disadvantage and the answer is reporting that shows the work rather than a different pricing model.
The commercial terms. The same as everything else we do.
Month to month, no minimum term, no setup fee. Accounts, analytics configuration and content live in your name, so if you leave you leave with all of it. Our cost page covers the pricing question in full.
Reporting And Review
Three rhythms, all of them fixed rather than offered on request. The cadence is the same across every service we run.
A written update every three weeks. What was done, not what is planned.
Short, specific and about completed work. Three weeks rather than monthly because it means a problem surfaces before a month has been paid for.
Monthly reporting. The five inputs from block five, each with its limitation.
Including the ones that did not move. Reporting that only contains favourable figures is managing you rather than informing you.
A full technical audit quarterly. Including crawler access re-checked.
Hosting changes, security rules and platform updates all break access silently, so this is checked again rather than assumed to be holding.
What the quarterly review also covers. The plan itself.
Which items were bets, whether any assumption behind them has changed and whether the sequence still makes sense. Our strategy guide explains why recording the assumptions is what makes review possible.
What we would raise unprompted. Two things.
If the constraint turns out not to be search at all. And if we reach a point where we think the money is better spent elsewhere, which we would say rather than continue billing.
What Changes When The Field Changes
This subject moves faster than anything else we work in, so a position taken now will need revising. The commitment is to revise it rather than defend it, then to tell you when we have.
What would trigger a change. Four things.
A vendor publishing something that turns an observation into documented behaviour. A vendor changing or withdrawing a control, which has happened more than once. A shift in which products your customers actually use. Or your own sampling moving consistently across quarters.
What would not. Rather more.
One month of sampling moving. An article asserting a new ranking factor. A competitor appearing in a single answer. None of those is evidence. Treating them as such is how a plan becomes a reaction.
How you would hear about it. In the quarterly review, in writing.
With what changed, what we had previously assumed and what we are doing differently. Including where the change means less work rather than more.
What happens to these pages. They get corrected.
The four platform guides and the AI Overviews guide carry monthly review dates. The reasoning pages carry quarterly ones. Every vendor claim in this cluster is attributed with the date it was checked, so a reader can see how current it is rather than guessing.
And if we turn out to have been wrong. We will say that too.
A position published with a date and a source can be corrected. One published as timeless cannot, which is why we date everything.
Restraint is the only
differentiator worth having.
No guaranteed appearance, because nobody controls it. No proprietary score, because we would have to invent the method. No statement about internal mechanics nobody has published. No timescale we cannot support.
What is included every month:
One target area. No setup fee, no minimum term, 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.
