The AI Overview quoted a forum thread and a competitor we had never outranked, while our position-three listing sat unused underneath.
That is the split. Ordinary SEO still fights for the blue link. AI SEO has to earn a citation inside the generated block, or the click never happens. The two jobs share crawl and entities. They diverge on what winning looks like on the results page.
When a marketing lead asks us for AI SEO, the request is usually this: keep the rankings already paid for, and stop disappearing when Google answers the query itself.
What ordinary SEO still has to do
Titles, internal links, indexation, and a crawlable HTML body still decide whether a page exists as a candidate. Nothing in AI search repeals that.
If Googlebot cannot fetch the passage, the Overview cannot cite it. If the canonical is a mess, the wrong URL gets the mention. If the sitemap is stale, new answers stay invisible.
Ordinary SEO is the floor. We still map templates, fix robots rules, compress the critical path, and stop faceted URLs from eating crawl budget. That work did not become optional because a generated paragraph appeared above position one.
What changed is the prize. A stable rank-three listing used to be a traffic plan. Now it can sit under a block that answers the query and sends the user away.
Where AI Overviews actually pull from
Overviews are assembled from several sources, not from the first organic result alone. A page that never ranked in the top three can still be cited if the passage is clear, attributed, and easy to lift.
A page that ranks well can be ignored if the answer is buried under hero copy, a cookie wall, or a JavaScript shell that never rendered the facts.
Google documents how these blocks appear, and what site owners can expect, in the Google AI features guidance. That is the baseline we use when a brief asks what still counts as a ranking and what now counts as a citation.
The sentence after that page is the practical one: there is no separate “AI ranking factor” we can buy. The systems that select supporting links still start from crawlable, people-first pages. The Overview is a new surface on top of that selection, not a replacement index.
Citations are not a ranking
A citation is a named source inside the generated answer. A ranking is a position in the classic list. Mixing the two in a dashboard is how a marketing lead ends up celebrating a blue-link hold while branded traffic falls.
We track both. Rank for the money queries. Citation for the queries where Google now writes the answer. Share of voice for the entity: does the Overview mention our client’s brand, product, or definition, or does it mention a competitor and a Reddit thread.
If the Overview cites us and the click still drops, that is not a failure of the citation. That is a zero-click query. The next useful move is to win the brand mention, then give the cited page a reason to click anyway: a calculator, a table, a local proof point, a download the model cannot paste.
If the Overview never cites us and rank is fine, ordinary SEO is not finished. The page is visible to the list and invisible to the block. That is an extraction problem, not a backlink problem.
Entities over strings
Ordinary keyword work still matters for matching. Entity work decides whether the model trusts the page as a source.
We mean the organisation, the product, the author, the place, and the relationships between them, written the same way on the page, in the schema, in the about copy, and in the knowledge sources Google already has.
A PH site that calls itself three different legal names, with no Organisation markup and an author who is only a first name, is asking the model to guess. Models do not guess in our favour. They cite the source with a stable entity.
This is where a lot of “just write more blogs” briefs die. Another 1,200 words on a synonym does not fix a missing entity. A consistent NAP, a clear sameAs graph, and a definition paragraph the model can lift will do more for an Overview than a sixth listicle on the same head term.
Write the entity in the first screen. Repeat it the same way. Link it to the pages that prove it. That is AI SEO that still looks like ordinary on-page, done with less slop.
Crawl still gates the answer
If the HTML the crawler receives does not contain the answer, there is nothing to cite. Client-side rendering, blocked CSS that hides the main text, infinite scroll without a paginated fallback, and soft-404s on filtered URLs are still the usual killers.
We see this on Philippines builds that ship a React shell and put the commercial copy behind a fetch. Ordinary SEO already hated that. AI SEO hates it more, because the Overview wants a clean passage, not a loading state.
This is why technical SEO services Philippines sit in the same journey as the AI brief. Crawl, render, canonical, and status codes are not a separate product story. They are the reason a well-written answer never appears in the block.
Fix robots and render first. Then argue about tone. A beautiful answer that Googlebot never saw is ordinary SEO debt with a new name.
On-page that a model can quote
Ordinary on-page packed a keyword into the title, H1, and first paragraph. AI on-page still does that, then adds a passage the model can lift without rewriting the whole page.
That passage is usually a direct answer, a definition, a numbered process, or a small table of facts with units. It sits high. It does not open with a weather report about the industry. It names the thing, states the claim, and points to the proof.
We write for extraction without writing like a FAQ dump. One clear claim per section. Evidence under it. No throat-clearing. No “in today’s digital landscape.” The model skips that filler and so should we.
Unique data helps. A method, a sample, a constraint from a real migration or a real audit is harder for the Overview to replace with a generic competitor paragraph. Invented statistics do the opposite: they get ignored or they get us into a helpful-content problem.
If the page cannot be quoted in two sentences, it is not ready for AI search. It may still rank. It will not be cited.
Measurement ordinary rank trackers miss
Rank tracking still has a job. It does not tell us whether the Overview appeared, who was cited, or whether our URL was a supporting link.
We add three views a marketing lead can actually use. Overview presence on the money queries. Citation of our domain versus competitors. Click and assisted conversion on the URLs that do get cited, because a citation with no landing-page job is vanity.
Search Console will not hand this over as a single “AI SEO” column. We pair query reports with manual SERP captures on the terms that matter, then we watch branded demand. If the Overview names us and branded search rises, the block is working even when the click on that query falls.
If rank is flat, Overview presence is zero, and branded search is flat, the AI brief has not started. We are looking at ordinary SEO that has not yet produced a citable passage.
Do not kill a page because clicks dropped on an Overview-heavy query if citation and brand both moved. Do kill a page that never gets cited and never ranks. That page is not in the fight.
Technical work AI search still needs
The stack did not change because the SERP grew a generated paragraph. Core Web Vitals still affect crawl efficiency. Hreflang still stops the wrong country page being cited. HTTPS, clean status codes, and a coherent internal link graph still tell the systems which URL is the source.
What we add for AI search is ruthlessness about the text the crawler actually receives. View-source and the rendered DOM have to contain the claim. Images of text do not count. Accordion content that is absent from the HTML does not count. A PDF with the only accurate spec may count, but only if it is crawlable and linked from a HTML hub we control.
Log files still tell the truth. If Googlebot is burning budget on parameter URLs and skipping the definition pages, the Overview will keep citing someone else. That is not a content mood. That is a crawl budget meeting.
Schema is not a magic AI switch. It is a consistency layer for entities, products, FAQs, and organisations. We use it when the visible page already says the same thing. We do not stuff FAQ schema that the user cannot see.
What changes in the brief
An ordinary SEO brief says: rank for these terms, fix these templates, publish this cadence. An AI SEO brief says all of that, then names the queries where we must be the cited source, the passages that must be extractable, and the entity we are trying to make unambiguous.
That is the work behind AI search optimization services Philippines when a marketing lead asks for AI SEO instead of another content calendar. The calendar can stay. The unit of work becomes the citable passage and the entity, not the word count.
We also change the competitor set. Ordinary SEO compared the sites in the top ten. AI SEO adds whoever the Overview is already quoting: publishers, forums, government pages, and brands that never ranked in our vertical’s classic list. If the block loves a standards page, we do not try to out-blog it. We cite it, then add the operational detail that page will never have.
The brief should say what we will stop doing. Thin roundups that restate the SERP are fuel for someone else’s Overview. We cut them. The hours go into the pages that can be the source.
How we treat content versus development
Content without a crawlable template is a draft. Development without a passage to render is a shell. We run them as one job when the SERP is an Overview.
Migrations are the sharp version of this. A new stack that drops server-rendered product copy will lose citations in a week, even if rank-tracking tools still show a URL. We treat that as an AI SEO incident and an ordinary SEO incident at the same time, because it is both.
White-label and in-house teams hit the same wall when the CMS stores the answer in a module the crawler never sees. The fix is not a new keyword map. The fix is the HTML.
Local work still matters in the Philippines for maps and for entity place. An Overview that needs a city, a regulation, or a service area will prefer the page that states it plainly. Hide the area served in an image and we are back to ordinary local SEO debt.
Decide the next useful step
Start with five queries that used to send clicks and now show an Overview. Capture who is cited. Capture whether our URL is in the classic list. That screenshot is the brief.
If we are in the list and not in the block, the next step is extraction: a quotable passage, a cleaner entity, a render that includes the facts. If we are in neither, the next step is ordinary technical and on-page until the page is a candidate. If we are cited and clicks still fell, the next step is a reason to visit: a tool, a table, a local proof, a next action the generated text cannot replace.
When those five queries are honest, a scoped audit is worth requesting. We will say whether the gap is crawl, entity, passage, or measurement. We will not dress ordinary indexation debt up as a new product.
Bring the screenshots, the money queries, and the URLs that should have been the source. That is enough to decide if the fit is real.
