Skip to main content
AEO/GEO

GEO Content Strategy: Deciding What to Publish

July 2026 · 7 min read · Updated September 2026

Short answer. The hard part of a GEO content strategy is not how to write the page — it is deciding which pages are worth writing at all. The rule that holds up: publish where the question is commercially real, where the incumbent answer is weak or generic, and where you can say something non-substitutable — first-party data, a defined method, a stated formula, an honest limit. Everything else is a page a model can already assemble from three other sources, which is why it keeps using those three.

Key takeaways

  • Most GEO plans fail at prioritisation, not execution: writing the 40 easiest questions usually means writing the 40 already answered adequately elsewhere.
  • A question inventory should score four criteria — commercial reality, incumbent weakness, non-substitutability, and maintainability — each 0–3, before anyone writes a page.
  • The highest-value question sources are sales and support transcripts, direct testing of assistants, and in-market collection per language — not a keyword tool.
  • A significant share of "who should I use" and reputation questions are assembled from third-party platforms rather than vendor websites, so owned content alone cannot win them.
  • Fewer, maintained pages outperform a large library of unmaintained ones, because stale, checkable claims are a liability once an assistant repeats them.

What should a GEO strategy actually optimise for?

"Get cited by ChatGPT" is not a single objective; it collapses six separable outcomes into one, and treating them as one blended score makes diagnosis impossible.

Outcome What it means What moves it
Discoverability The page is in the index at all Crawlability, rendering, internal links
Retrieval The page enters the candidate pool for a question Topical match to the sub-questions actually asked
Selection The passage enters the model's context Self-containment, directness, evidence
Citation The source is visibly attributed Entity resolution, unambiguous ownership of the claim
Mention The brand is named without a link Third-party corroboration as much as owned content
Business result Pipeline, not visibility Whether the question was commercially real

Retrieval is the page entering the candidate pool for a question at all, before selection or citation can happen. A programme that only tracks one blended visibility number cannot tell "we are not in the index" apart from "we are in the index and boring," and those two problems need opposite fixes. A technical AEO checklist addresses the discoverability layer directly.

Where do the questions come from?

Not from a keyword tool, and not from an English question list translated into other markets. Three sources carry most of the value, in descending order.

  • Sales and support transcripts. The questions a buyer asks a human are close to the questions they ask an assistant. This is the highest-yield input, and the one most teams already own and rarely read.
  • The assistants themselves. Ask category, comparison and problem questions across engines and record which produce confident answers, which produce hedged ones, and who gets named — a practice covered in more depth in how people actually prompt AI assistants.
  • In-market collection per language. Questions differ between markets in substance, not only in wording. A translated list carries the source market's assumptions about what buyers care about, and the translated pages then underperform for reasons that have nothing to do with the writing.

Record each question with the market, the language, the buying stage, and who currently answers it well — that last field is what drives the publish decision.

How do you score a question?

Four criteria, each scored 0–3, decide whether a question gets a page and how much effort it gets.

Criterion 0 3
Commercial reality Nobody who asks this buys anything The question sits directly before a purchase decision
Incumbent weakness Answered well by an authoritative source Answers are generic, contradictory, or visibly hedged
Non-substitutability We would be restating public knowledge We hold data, a method, or an outcome nobody else can state
Maintainability The answer changes monthly and nobody owns it Stable, or a named owner exists to update it

A score of 10–12 gets full treatment with first-party evidence and a maintenance owner; 7–9 gets written once the tier above is complete; 4–6 becomes a section folded into an existing page rather than its own URL; 0–3 does not get written, and this bottom tier is what consumes most GEO budgets. The non-substitutability criterion does most of the discrimination: if the honest answer to "what can we say here that another source cannot" is nothing, the page will be correct, competent and unused.

What makes a page non-substitutable?

A model can already produce a fluent overview of almost any topic; what it cannot produce is a specific, attributable fact that exists in exactly one place.

Five things qualify. First-party measurement — numbers produced with the method stated, since a figure with a described method is quotable and a figure without one is only a claim. A defined formula, since passages that state how something is calculated are rare in most categories' body copy. A stated threshold, since "pass at eight per thousand words" is liftable and "high quality" is not. A named limitation, since pages explaining where a method fails read as more credible than pages claiming universal success. And an outcome with its conditions attached — what changed, over what period, measured how, and what did not move, rather than an unqualified "we improved results."

Google's own guidance to publishers on optimizing for generative AI features points the same way: unique, people-first content rather than near-duplicate pages generated for every query variant. The complete guide to generative engine optimization sets out how that principle extends across a full programme, not just individual pages.

What about the questions you cannot win on your own site?

A meaningful share of the answers in any category are assembled from third-party platforms rather than vendor websites, so publishing harder on your own domain only addresses part of the problem.

Owned content tends to win definitional, methodological and "how do I do X" questions, where the best available answer can genuinely be the vendor's own. Third-party corroboration tends to win "who should I use," "what are the alternatives to X," and reputation questions, which are assembled from review platforms, editorial listicles, community threads and reference sites — the dynamic examined in why third-party brand mentions matter for GEO. Neither channel wins quickly on model memory: being described accurately by a model answering with no browsing changes only at training cadence, through what exists about a brand elsewhere, over months. Earned coverage has to be genuine — planted reviews and citation farms produce short-term mentions, conflict with platform policy, and are the kind of signal that gets discounted once detected.

How should a publishing cadence work?

A cadence that survives contact with reality favours fewer, maintained pages over a large, aging library, and treats each page as owned rather than published and forgotten.

  • Fewer, maintained. A library of 25 pages that are kept current beats 120 that are not, because staleness is visible and dated claims are checkable.
  • One page per question, not per phrasing. Consolidating wording variants into one substantive answer avoids near-duplicates that compete with each other and trip scaled-content policy — the same discipline described in question headings and answer-first writing.
  • Name an owner per page. Unowned pages decay into wrong pages, and a wrong page is worse than an absent one once an assistant repeats it back to a prospect.
  • Re-score annually. Incumbent weakness is the criterion that changes fastest; a question that was wide open last year may now be answered well by someone with more authority. Content refresh operations for AI search covers how to run that re-scoring as a repeatable process.

How does Lifewood approach this?

Lifewood treats question selection as the first deliverable of a GEO programme, before any writing, and delivers a "do not write" tier explicitly rather than quietly.

Generative engine optimization (GEO) is the practice of shaping content so it is retrieved and used inside AI-generated answers, as distinct from ranking in a list of links. The question inventory is built from client sales and support material and from live assistant runs rather than keyword exports, scored on the four criteria above. Content is then produced in-market rather than translated: with 100+ languages and 40+ delivery centres across 30+ countries, the question inventory for each market is collected by people who sell into it, which is how the substance of a question survives the crossing between languages. See GEO services and AEO services for how that inventory work connects to delivery. The limit worth stating: none of this makes a page citable if the entity behind it cannot be resolved or the page cannot be fetched in full — those are fixed first.

Frequently asked questions

They share a foundation and diverge at the outcome. Both require crawlable, indexable, useful pages, but SEO measures position in a ranked list while GEO measures whether a passage is used inside a generated answer, where there is often no click. The prioritisation differs too: GEO rewards being the only source for a specific fact more than it rewards comprehensive topic coverage.

No. Consolidate related questions into substantial pages and decline the ones with nothing non-substitutable to say. Scaled near-duplicate content built mainly to catch query variants is explicitly discouraged by search guidance, and it competes with itself for retrieval rather than helping it.

Fewer than most plans assume, and maintained. The binding constraint is not page count but whether each page carries something a model cannot assemble elsewhere. Twenty-five pages with first-party evidence tend to outperform a hundred that summarise public knowledge, and the hundred also cost more to keep accurate.

Yes, but only as one input. Links remain part of how pages are discovered and how web authority is assessed, so they are not irrelevant. Citation behaviour, though, draws on a broader mixture of sources than a link graph describes, and a large share of category answers is assembled from third-party platforms rather than vendor sites.

Diagnose which stage is failing before writing anything. If competitors are named from third-party sources, more owned content will not fix it and the work is placement and corroboration instead. If they are named from their own pages, compare passages directly for directness, specificity, currency and attribution — rewriting against a known winner is cheaper than publishing more.

Baseline a fixed prompt set before publishing, then re-run it on a stable cadence, reporting retrieval and memory separately per market. Expect retrieval to move first, within weeks; if it does not move at all, check delivery before rewriting content, since most "the content did not work" outcomes are pages an engine never received in full.

Sources and further reading

  1. Optimizing your website for generative AI features — Google Search Central
  2. Aggarwal et al., "GEO: Generative Engine Optimization," ACM SIGKDD 2024 (arXiv:2311.09735)

Have an AI or visibility project in mind?

From AI evaluation and human-in-the-loop review to GEO and AEO strategy, our team can help you deploy with confidence and get found in the AI search era.

Talk to our team