SEO and GEO Optimisation

Ongoing SEO management starts with the questions real buyers use, the pages search systems return and the actions visitors take afterwards. We segment conventional search by query, landing page, country, device and search appearance where the platform supplies those dimensions. Brand demand is separated from non-brand demand. A high average position on a founder’s name cannot be used to disguise the absence of commercial queries, just as a growing impression total cannot by itself prove qualified reach.

AI-search observation is recorded in a different layer. A brand can be mentioned without a source link, cited without being recommended, recommended without producing a visit, or visited without producing an enquiry. We therefore log the prompt family, locale, platform, date, response state and visible source path when reproducible observation is possible. Search Console’s generative AI report, referral analytics and controlled answer checks can contribute evidence, but none provides a complete census of every answer shown to every user.

A weak result can begin at discovery, crawl access, rendering, canonical selection, indexing, relevance, authority, presentation or conversion. Those are separate failures. We inspect the maintained source and the public response, then reconcile redirects, robots controls, status codes, canonical tags, hreflang, sitemap membership, rendered copy, structured data, internal links and page experience. Only after the path is technically coherent do we ask whether the page offers enough original knowledge and evidence to deserve the query.

Content gaps are treated as knowledge and decision gaps, not empty keyword slots. The useful question is not how many URLs a competitor owns, but which page resolves a distinct need with clearer evidence. A country or city deserves a dedicated page only when the offer, language, regulations, examples, availability or buying conditions genuinely change. Replacing one place name across hundreds of near-identical pages would create doorway inventory and weaken the exact authority we are trying to build.

The monthly backlog is scored by commercial relevance, verified demand, present visibility, affected route family, confidence in the diagnosis, implementation effort and regression risk. We favour changes that clarify an existing canonical page, repair a systemic template fault or connect an evidence cluster before commissioning another broad article. A query at position eighty with the wrong landing page may need a different intervention from a query at position eleven whose title and decision evidence are weak.

Each chosen item receives a hypothesis and an acceptance test. The hypothesis states which observable layer should change and why. The acceptance test states what we can prove at release: a correct redirect, a self-canonical, reciprocal hreflang, an accessible answer, a stronger evidence path, a cleaner internal route, a stable layout or a measured conversion event. A later increase in impressions, ranking, AI exposure or enquiries remains an outcome to observe rather than a condition we pretend to control.

Implementation takes place in the canonical source and within the capacity agreed for the month. Technical work may cover routing, indexing controls, templates, performance, structured data, language relationships, sitemaps and analytics. Editorial work can reshape a service page, comparison, answer, guide or evidence record around the decision it must serve. Entity work aligns the organisation, named experts, methods, publications and verifiable claims across visible copy and machine-readable records without inventing offices, customers, reviews or awards.

For multilingual search, facts and commercial terms remain equal while each locale is written for its own readers. Query vocabulary, sentence rhythm, examples and calls to action are rebuilt in the language rather than translated mechanically. Arabic is checked in an RTL layout and mixed-direction strings are tested. Reciprocal hreflang and self-canonicals connect equivalent pages, but language tags do not compensate for thin or unnatural copy.

Before deployment we test the source, generated routes, metadata, structured data, internal links, language parity, responsive layout and accessibility. A bounded release manifest preserves the intended scope and a recoverable backup. After deployment we fetch the public URLs independently, compare their bytes or required semantics, sample ordinary and named crawler access, and verify the conversion route. Upload success is not public verification; public verification is not indexing; indexing is not ranking.

Changed routes enter a dated cohort so later Search Console and analytics comparisons use an eligible observation window. The monthly review keeps crawl, index selection, impressions, clicks, position, generative AI exposure, third-party answer observations, referrals, enquiries and revenue as separate lines. Movement is compared by query family, page and market where the data permits. When the evidence contradicts the hypothesis, we change the diagnosis rather than protect the report.