The last pitch promised twelve citations in Gemini by June. It named no source pages, no entities, and no answer-ready copy we could ship.
You are comparing generative engine optimisation and answer engine optimisation decks. The labels are new. The failure is old: a count with no URL behind it, sold as if a model were a media buy.
We work British-led SEO in the Philippines - technical, on-page, content, local, audits, migrations, white-label, development, and AI search. When a GEO or AEO slide lands, we do not score the slogan. We score whether the work produces pages an engine can quote without inventing a fact.
What those decks actually sold
Most of the decks we see this year collapse three jobs into one line. Generative engine optimisation. Answer engine optimisation. “Citations.”
The first job is retrieval: can a model find a page, trust it, and use it when it writes an answer. The second is extraction: can an answer engine lift a clean span without rewriting your meaning. The third is not a job. It is a screenshot someone hopes will appear.
A citation is an outcome of retrieval plus extraction plus whatever the model does that week. Nobody on our side of the table controls the third part. A pitch that leads with a citation quota is selling the part it cannot do.
Ask for the source pages. If the deck cannot name them, the rest is theatre.
Generative engine optimisation is not a citation quota
Generative engine optimisation is the work of making your site the material a generative system can retrieve and lean on. Training cuts, browsing, grounding, retrieval-augmented generation - the plumbing changes. The input does not. Models still need pages with clear entities, stable facts, and copy that answers a question in language a human would actually use.
We do not buy a slot in ChatGPT. We do not lease a mention in Perplexity. We build and repair the pages those systems fetch when they bother to fetch the open web.
That means indexable HTML, not a wall of client-side render. It means one canonical URL per fact, not five near-duplicates fighting. It means the organisation, the product, the people, and the places are named the same way on every source page, because a model that sees three spellings will pick a competitor that only used one.
If a vendor cannot walk a URL, an entity, and a paragraph they would be willing to see quoted, they are not doing generative engine optimisation. They are selling a mood.
Answer engine optimisation starts with extractable pages
Answer engine optimisation is narrower and older than the acronym. Featured snippets, People Also Ask, AI Overviews, voice answers, and a growing set of chatbot retrievals all prefer a span they can lift.
Lift fails when the answer is buried in a 2,400-word overture. Lift fails when the H2 is cute and the first sentence is a throat-clear. Lift fails when the same question is answered three ways on three URLs and the engine has to guess which one you meant.
Answer-ready copy leads with the answer, then the proof, then the edge cases. It uses the question the buyer typed, not the slogan the brand prefers. It keeps units, dates, names, and jurisdictions consistent so a model does not have to “help.”
AEO is not a schema plugin and a hope. Schema helps machines find the block. The block still has to be true, short enough to extract, and unique enough to be worth extracting.
Source pages we can stand behind
A source page is a URL we would allow an engine to quote. Not a tag archive. Not a thin location page with the city name swapped. Not a press mention we do not control.
For GEO and AEO, we inventory source pages the way a technical audit inventories index bloat. What facts live here. Which entity does this URL own. What question does the first screen actually answer. What would break if this URL were the only thing a model saw.
If the important facts live in PDFs, decks, and staff heads, we do not have source pages. We have folklore. Generative systems do not cite folklore. They invent a sentence that sounds like it, or they cite the competitor who wrote it down.
The useful next move on a weak inventory is not “more blogs.” It is deciding which URLs are allowed to be the record, then making those URLs complete.
Entities that survive a model rewrite
Entities are the nouns a system can hang a fact on: the company, the service line, the people, the cities, the regulations, the products. If those nouns drift, the model rewrites you into a neighbour.
British-led work delivered from the Philippines is a live example. UK English, PH English, legal names, trading names, and “SEO Philippines” versus a specific service all compete. If about pages, service pages, and bios disagree on who we are and where the work is done, retrieval gets noisy. Noisy retrieval is how a pitch’s “citation” becomes someone else’s brand.
We map entities before we brief copy. Same legal name. Same service labels. Same geo where it matters. Same people with the same roles. Same third-party profiles pointing at the same source pages. That is not brand fussiness. It is how a generative system decides you are one thing instead of four.
If a GEO pitch never mentions entities, it has not started.
Answer-ready copy versus a content calendar
A calendar of “thought leadership” is almost never answer-ready. Engines do not need another 1,800 words on why AI search matters. They need the definition, the steps, the comparison, the limit, and the date the limit was true.
Answer-ready copy looks uneven on purpose. Short where the answer is short. Long where the proof has to be long. A table where a table is the honest form. A “no” where the honest answer is no.
We write for the question the buyer is comparing, because that is the query generative systems get. “What is generative engine optimisation” and “how is AEO different from SEO” are not thought-leadership themes. They are extraction targets. If our source pages dodge them, a model will fill the gap from whoever did not dodge.
Filler also trains the wrong pattern. Thin, scaled, interchangeable pages are exactly what Google has been telling the industry to stop. The Google helpful content guidance is still the bar we use before we talk about generative engine optimisation or answer engine optimisation, because systems that quote the web still prefer pages written for people with experience of the work.
How we kill a fake GEO pitch in one call
We ask five things. The answers tell us whether to stay in the room.
Which source pages will you create or repair in the first ninety days, by URL pattern, not by theme. Which entities will those pages own, and where do those entities currently conflict. Which questions will the copy answer in the first two sentences. How will you know a page is extractable - not “traffic,” not “citations,” but a span a human can highlight. What happens when the model does not cite us anyway.
If the vendor answers with a tool stack, a prompt library, and a promised mention count, the pitch is fake. Tools can help crawl, cluster, and watch SERPs. They cannot obligate Gemini.
If they answer with pages, entities, and answer-ready copy, we can scope. That is the whole filter.
Content work that engines can quote
Content for GEO and AEO is not a separate product line we sprinkle on after rankings. It is the record. Definitions, comparisons, process, limits, and local fact sheets that a model can reuse without lying.
Our content SEO services Philippines work is built that way: source pages first, entities named, answers in the open, proof underneath. We do not brief “a blog about AEO.” We brief the questions a buyer comparing pitches will type, then we make one URL the honest answer.
That also means saying what we will not write. We will not manufacture experience we do not have. We will not spin the same paragraph across cities. We will not hide the answer below a brand story so the page “feels premium.” Premium that cannot be extracted is invisible to an answer engine.
When the inventory is thin, we cut before we add. Ten extractable source pages beat forty posts that all say “it depends” in paragraph six.
On-page work that makes answers cheap to extract
On-page is where AEO either happens or dies. Title, H1, first paragraph, subheads, lists, tables, internal links, canonical, language, and the HTML the bot actually gets.
If the answer lives in an image, it does not live. If the important sentence is in a tab that never renders for Googlebot, it does not live. If every H2 is a pun, extraction has to guess. Guessing is how you get a confident wrong answer with your brand nearby.
Our on-page SEO services Philippines work is the unglamorous half of answer engine optimisation: one H1 that matches the question, an opening span that could stand alone, headings that name the next useful thing, and markup that describes what is already on the page instead of hoping schema will invent it.
We also check the boring access questions. Can the page be crawled. Is it canonical. Is it duplicated on parameters. Does a soft 404 look like a real answer. Generative retrieval that hits a junk URL will either skip you or quote junk. Both are expensive.
What the scoped audit has to show
A scoped audit for GEO and AEO is not a 90-page crawl dump. It has to answer the buyer’s actual decision: is there enough on this site to be worth retrieving, and if not, what is the first set of source pages to build.
We want a list of current URLs that could serve as source pages, with the entity each one owns. We want the conflicts - two services, two names, two geos, two answers. We want the questions the copy already answers in the first screen, and the questions it dodges. We want technical blocks: render, robots, canonicals, language, duplication.
We want a gap list that is small enough to ship. Not “own AI search.” Own five questions with five URLs and copy we would let a model quote tomorrow.
If the audit cannot produce that list, it is not an audit. It is a pitch with extra PDF.
The next useful step
Take the decks in front of you and strike every citation number. What remains should be source pages, entities, and answer-ready copy. If nothing remains, that vendor is done.
Then look at your own site the same way. Pick one question a serious buyer would ask. Find the URL that should answer it. Read the first seven lines. If you would not let an engine quote those lines, that is the job - not a new acronym, not a prompt pack.
When that gap is real, and the work is pages we can stand behind rather than mentions we cannot control, request a scoped audit. We will say if the inventory is too thin, if the entities are a mess, or if the next useful thing is a handful of extractable URLs instead of a GEO theatre tour.
