LEARN
GEO & AI Search: Concepts and Methods
How generative engines pick what to cite, and what actually moves brand visibility. One concept per page, one clear position.
65 pages
What GEO Is
GEO is the practice of organising content and evidence around real decision questions so a brand is easier to include and recommend inside a generative engine's answer.
Updated 2026-08-10GEO versus SEO
GEO and SEO share most of their technical foundation; the real differences are the target form, the measurement, and the shape of the deliverables.
Updated 2026-08-10Are GEO, AEO and LLMO the Same Thing
GEO, AEO and LLMO describe the same work and differ only in emphasis and origin; the label does not change what needs doing.
Updated 2026-08-10How an AI Picks an Answer
Content influences two steps — being retrieved and surviving reranking; reading intent and choosing wording are outside your control.
Updated 2026-08-10Why an AI Cites One Page and Not Another
Source type sets the base probability of being cited; how you write decides who is chosen among similar sources — two separate investments.
Updated 2026-08-10What Zero-Click Does and Does Not Mean
Rising zero-click changes the relationship between being seen and being clicked; it does not automatically mean traffic falls — but measurement has to change with it.
Updated 2026-08-10Widely Repeated GEO Claims That Lack Evidence
In a field where the mechanism is not public, whether a practice is worth doing depends on checkable evidence — not on how many people repeat it.
Updated 2026-08-10The Actual State of This Field
Real demand, small volume, unsettled methodology — all three hold at once, and together they set the sensible pace of investment.
Updated 2026-08-10Fix Crawlability Before Content
AI visibility diagnosis runs four layers in order — crawlable, retrieved, mentioned and cited, converting — and while the crawl layer is broken, content work returns nothing.
Updated 2026-08-10Decide Retrieval and Training Crawlers Separately
Retrieval crawlers decide whether you can be cited; training crawlers decide whether your content trains the model — decide them separately rather than with one blanket rule.
Updated 2026-08-10Verify Crawling With Server Logs
AI-side crawling is effectively invisible in web analytics; server logs are the only first-party source for whether content was fetched and what was returned.
Updated 2026-08-10User-Agent-Based Prerendering Is a Trap
User-agent-based prerendering serves AI retrieval crawlers an empty shell while the page renders perfectly in a browser — only comparing responses across agents reveals it.
Updated 2026-08-10Do Not Put Key Content Behind Client-Side Rendering
A crawler reads the HTML you return; content and structured data injected by scripts do not exist for crawlers that do not execute JavaScript.
Updated 2026-08-10CDN-Level Blocks Are Not in Your robots.txt
robots.txt is a declaration and the CDN is a network-level switch; when they disagree, the network wins.
Updated 2026-08-10Structured Data Is a Floor, Not a Lever
For AI search, structured data is a floor rather than a lever: have it, but the difference does not come from it.
Updated 2026-08-10The Three Things With Evidence Behind Them
Three writing methods have evidence behind them — adding statistics, citing sources, adding quotations — worth a 30–40% relative visibility lift across 10,000 queries.
Updated 2026-08-10Put the Conclusion First
The sentences that get used are the ones that hold up out of context — put the conclusion first, not after the reasoning.
Updated 2026-08-10Original Data Is the Highest-Return Content
Original data solves three things at once: the most direct route to citation, the reason media will cover you, and the only reliable way to attribute an answer to a specific page.
Updated 2026-08-10Organise Pages by Passage, Not by Chapter
A page does not need to answer the whole question — being the better answer to one sub-query is enough to be used.
Updated 2026-08-10One Position Per Page
What gets cited is a judgement that can be lifted directly; a page arguing both sides offers neither a quotable conclusion nor an action.
Updated 2026-08-10Designing a Content Cluster
A content cluster exists so that every sub-query produced by query fan-out lands on a page written to answer it.
Updated 2026-08-10Long-Tail First — Not Because Head Terms Are Impossible
Long-tail comes first not because head terms are impossible, but because head terms require authority built from pages, links and citations.
Updated 2026-08-10Compute the Page Matrix Before Setting Volume
Page count is derived — enumerable objects × page types per object gives the ceiling, rather than setting a volume and hunting for topics.
Updated 2026-08-10A Glossary Is the Foundation for Citation
A one-sentence definition is the most liftable form of content, which makes a glossary the foundation for citation rather than a secondary column.
Updated 2026-08-10Thin Pages Cost You Twice
Thin pages are not merely useless: they fail to rank while diluting crawl budget and internal authority.
Updated 2026-08-10Internal Links Are Designed, Not Added Ad Hoc
Internal links solve two things at once: reachability from any entry point, and expressing how concepts relate.
Updated 2026-08-10Most Citation Weight Is Not on Your Site
Roughly three quarters of citation weight sits off-site, so owned content alone hits a structural ceiling — not a quality one.
Updated 2026-08-10The Entity Signals Checklist
Entity recognition rests on three signals — your own declarations, third-party records, and consistency across the web; when they conflict, models describe you vaguely or wrongly.
Updated 2026-08-10Unify the Description Before Distributing It
When descriptions disagree, a model has no basis to choose and produces vagueness — unify first, distribute second.
Updated 2026-08-10Communities Need Genuine Participation, Not Placement
Communities are the highest-weight source type and the least suited to a placement approach — promotional posting gets identified and undermines the credibility you are trying to build.
Updated 2026-08-10B2B Review Sites Are Worth Setting Up
Review sites are cited heavily on B2B decision queries because those questions want a third-party view, and that is precisely what a review profile provides.
Updated 2026-08-10Digital PR Needs Something Reportable
What media picks up is rarely an opinion and usually a number — everyone has opinions, and a number is something nobody else has.
Updated 2026-08-10Structured Knowledge Bases Before Encyclopedia Entries
Structured knowledge bases have a far lower bar than encyclopedia entries, and an entry created before independent coverage exists gets deleted.
Updated 2026-08-10Bylined Content and Personal Credibility
A byline ties content to a verifiable person and makes 'who is saying this' checkable — which is easier to believe than institutional voice.
Updated 2026-08-10The Reality of Distribution
Distribution can commit inputs — which channels and to what standard; whether it is run, kept or cited is not ours to decide.
Updated 2026-08-10The Chinese and English Ecosystems Cite Different Sources
The Chinese and English citation mixes are two different sets; porting the English list across means investing where citations do not happen.
Updated 2026-08-10Chinese AI Search Starts With Ecosystem Ownership
On the Chinese side, AI platforms and content platforms often share an owner, which gives citation preference a visible ecosystem bias.
Updated 2026-08-10What Each Chinese Content Platform Is For
Chinese content platforms are not interchangeable — different readers, different citing AIs, different suitable formats — so one piece everywhere is waste.
Updated 2026-08-10How Cross-Border Tech Companies Should Prioritise
Going global is not about switching to English — the citation sources, the measurement and the buyer's basis for trust all change.
Updated 2026-08-10Localisation Is Not Translation
Translated content usually earns citations on neither side — the phrasing, the priorities and the citable evidence are all different.
Updated 2026-08-10Visibility for Overseas Companies Entering China
Overseas companies face three structural differences in China: sources belong to different ecosystems, most platforms hide sources, and content expectations differ.
Updated 2026-08-10Measurement Discipline: Four Rules
Four measurement rules: fix the denominator, never sum different sources, report per platform and language, and label estimates as estimates.
Updated 2026-08-10First-Party Data First
AI visibility attribution runs on first-party analytics and server logs; requests that do not execute JavaScript never reach front-end analytics at all.
Updated 2026-08-10Tier the Commitments
Commitments come in two tiers: a pilot commits to a quantified lift in mentions, and only a full engagement commits to measurable organic growth.
Updated 2026-08-10What Not to Promise
Four things not to promise: a specific rank, post-visit conversion, third-party platform output, and targets built on third-party estimates.
Updated 2026-08-10Measure the Baseline Before Touching Anything
Without a baseline, no later change can be attributed — you cannot separate your work from market drift.
Updated 2026-08-10Position Bands Decide Priority
Beyond position 11, a keyword returns about 0.1 clicks a month — nearly all yield in this category sits in the top ten.
Updated 2026-08-10Publish in Stages and Validate the Template First
Staged publishing is not about limiting output — it stops an unvalidated page type from being replicated two hundred times.
Updated 2026-08-10A Review Has to Cut Something
A review's value is deciding what continues and what stops; an additive-only review keeps funding directions that return nothing.
Updated 2026-08-10Building a GEO Strategy From Scratch
Five steps: check the crawl layer, build the question set and baseline, compute the page matrix, publish in stages, review by band — and the order is fixed.
Updated 2026-08-10What to Do in the First 90 Days
The first 90 days: two weeks of groundwork, six of ramp, four of review — ramping early contaminates the feedback.
Updated 2026-08-10How to Build a Question Pool
A question pool is not a keyword list: collapse surface phrasings into decisions, then split into a frozen monitoring pool and a growing topic pool.
Updated 2026-08-10Compute the Page Matrix Per Company
A page matrix is computed — enumerable objects × page types — and different companies arrive at very different counts.
Updated 2026-08-10Resourcing and Cadence
Production is usually not the constraint; topic space, technical groundwork and genuine off-site participation are.
Updated 2026-08-10An Action List for Marketing Leads
Separate what marketing can drive from what engineering must confirm first — without the second, the first can return nothing.
Updated 2026-08-10A Checklist You Can Actually Tick
The checklist runs four layers — crawlable, retrieved, content and evidence, conversion — and every item needs a verifiable test.
Updated 2026-08-10When This Is Not Worth Doing
Four situations where this does not pay: buyers do not decide inside AI, the product is unsettled, there is no checkable evidence, and nothing catches the visit.
Updated 2026-08-10When AI Answers Get Your Brand Wrong
AI describes brands wrongly for three reasons — stale, conflated, invented — and each has a different fix, so classify first.
Updated 2026-08-10Why AI Gets Brands Wrong
Most brand errors are not invention — the model retrieved the wrong thing or failed to retrieve the right thing, and all four causes are controllable.
Updated 2026-08-10A Correction Playbook
Correction is not a statement — while the wrong version stays retrievable, adding one document does not change the output.
Updated 2026-08-10Handling Negative Descriptions in AI Answers
Negative descriptions in AI answers split three ways — true, outdated, conflated — and what can be done differs completely.
Updated 2026-08-10The Boundary: Content and Evidence, Not Monitoring
The boundary is content and evidence: what you publish and whether it checks out — not web-wide monitoring, and not when a model changes its wording.
Updated 2026-08-10How to Judge Whether a Vendor's Claim Holds
Four questions decide whether a claim holds: the source, the sample size, whether a platform said the opposite, and whether the definition is checkable.
Updated 2026-08-10The Equaliser Effect: Who Gains Most
After applying these methods, pages ranked fifth gained 115.1% in visibility while first-ranked pages lost 30.3% on average — the return is higher for brands outside the top tier.
Updated 2026-08-10AI Traffic Is Small but Different in Kind
AI referrals are still around 1% of traffic and growing 357% year over year — presenting either figure alone distorts it.
Updated 2026-08-10