Search everywhere optimization
The practice of treating every surface where people ask questions — engines, assistants, marketplaces, video and community sites — as a retrieval target, rather than optimising one engine.
Karl-Gustav Kallasmaa, Founder & CEOLast updated Search everywhere optimization is the practice of treating every surface where people ask questions — search engines, AI assistants, marketplaces, video platforms, app stores and community sites — as a retrieval target, rather than optimising a single engine and hoping the rest follow. The phrase is trade vocabulary rather than a standard; what makes it useful is the shift in unit it forces, not the label.
Why the unit changes
Classical SEO optimises a document to occupy a position in a ranked list, and assumes the position produces a click. Most of the surfaces this term adds do not return a list. An assistant returns one composed answer. A marketplace returns a card. A video platform returns a result whose useful content is a spoken passage inside a file. In each case the thing selected is smaller than a page and is presented without the page around it.
That is the shift: from optimising a document for placement to optimising a passage for extraction. A section that reads correctly only in the context of the two sections above it will be quoted incorrectly, or not quoted at all.
The pressure behind the shift is measurable. A 2024 SparkToro and Datos study, drawn from a US and EU panel over September 2022 to May 2024, found that 58.5% of American Google searches resulted in zero clicks. When the majority of queries end on the answering surface, a strategy whose only success metric is a session is measuring a diminishing share of the outcome. Zero-click search covers this directly.
The mechanism per surface
Surfaces differ in how they retrieve, and the differences are concrete rather than stylistic:
- Classical engines crawl, index and rank documents. The lever is the index entry.
- AI assistants that fetch at query time perform an HTTP request during the answer and write from what returns. The lever is the passage, and the precondition is access. See AI search.
- Assistants answering from parametric memory use no live fetch at all. Nothing published this month reaches them; the only lever is what was already widely and consistently stated when the model was trained.
- Marketplaces, app stores and video platforms run closed internal indexes over structured fields you supply. The lever is the field, not the prose.
- Community sites are both a surface and a corpus: assistants retrieve their threads, so a correct answer written there can be quoted about you without your site being fetched at all.
Access is upstream of all of the fetching cases. OpenAI documents four separate agents doing four separate jobs — OAI-SearchBot for surfacing sites in ChatGPT's search features, OAI-AdsBot for ad safety validation, GPTBot for foundation-model training, and ChatGPT-User for page visits a user's question triggers. A site that added a blanket rule to opt out of training frequently opted out of retrieval as well, and the symptom is silence rather than an error. robots.txt covers how to answer each agent deliberately.
What to actually do
- Enumerate the surfaces your buyers use, not the surfaces with the largest total audience. A B2B tool answered inside an assistant does not benefit from a video platform strategy.
- Write for extraction on all of them. One claim per section; the qualifier in the same sentence as the claim; product names, version numbers and identifiers written out as literal text.
- Fill structured fields where the surface has them. A marketplace ranks on its schema, not your paragraphs.
- Make access explicit. Decide per agent, in
robots.txt, whether it may fetch — and check the decision matches what you believe it is. - Verify what each surface says about you, because the failure is usually a wrong claim rather than an absent one, and a wrong claim is traceable to a passage you can edit. GEO covers the measurement side.
Failure modes
- Treating it as a channel checklist. Being present on nine surfaces with unextractable prose performs worse than being correct on two.
- Assuming one crawl policy covers every agent. Training, search and user-triggered fetches are separate agents with separate purposes.
- Porting ranking tactics to a surface with no ranking. There is no position inside a generated paragraph to occupy.
- Ignoring third-party surfaces. A community thread is retrieved and quoted whether or not it is accurate; it is part of the corpus about you regardless of who wrote it.
- Measuring only sessions. On surfaces that answer in place, the visit is not the event; the correct mention is.
Frequently asked questions
How does it differ from SEO?
SEO places a document in a list. This optimises a passage for surfaces that return one answer.
Which surfaces count?
Any that retrieve and answer: engines, assistants, marketplaces, video, app stores, community sites.
Does one robots policy cover all agents?
No. Operators document distinct agents for search, training and user-triggered fetches.
What changes about the writing?
Sections must be correct when lifted out, with the qualifier in the same sentence as the claim.
Terms related to Search everywhere optimization
A search that ends without the reader visiting any website, and the measured gap between sessions that show an AI summary and those that do not.
Search where a model composes the answer and fetches pages through named crawlers, rather than returning a ranked list of links for you to read.
The practice of getting a source reproduced inside an AI-generated answer, introduced as a named paradigm in a 2023 research paper.
The standardised file that tells crawlers what they may fetch, now the main place where AI training and AI retrieval access are decided separately.