Home / Field guides / SEO & AI Search
Field guide · SEO & AI SearchSearch, Answers and the Evidence Beneath
A field guide to SEO and AI search for owners who intend to run some of it themselves.
Written to be used.
Some readers will finish this guide and decide to run the work themselves. That is a good outcome. Search rewards owners who understand their own market better than any outsider can, and a great deal of what follows is genuinely within reach of a careful person with modern tools and a few hours each week.
The guide is written to be useful on its own. It sets out what we would do first, what we would refuse to do, and the failures we are most often called in to repair. Where a decision turns on judgement rather than execution, we have tried to explain the reasoning, so that the judgement can be made without us.
You can generate the work. You cannot delegate the outcome.
What you can genuinely do yourself.
- Drafting and editing at volume
- A first map of demand
- Running and reading a technical crawl
- Your own titles and descriptions
- Consolidating before creating new pages
- Canonical form and URL structure
- Deciding what a new page displaces
- Redirect mapping through a redesign
- Arabic as a market of its own
If you have one language, one clear service, no inherited URL history and an owner with a few hours a week, you do not need us at all. Work through the six steps at the back of this guide in order, and keep going for two quarters without changing course.
Where the general answer stops being useful.
Not because it is wrong. Because it is general, and this is a contested problem against named competitors who were handed the same advice.
“Write 20 meta descriptions for my Dubai interior design page.”
Correct, usable, and near-identical in kind to what your competitors are being handed this week.
What the decision requiresA description is copy for the result, not for the page, and its job is to filter rather than to attract. The real decision is which visitor to turn away: the price shopper, the wrong emirate, the project two sizes too small. An enquiry you cannot profitably serve costs you more than the click was ever worth, and which enquiries those are is knowledge that sits in your own quotation history, not in the model.
“Give me a keyword plan for a private clinic in Dubai.”
A sound starting map, and better structured than most briefs we are sent.
What the decision requiresThe value of a plan is not the list, it is the assignment: which single page owns each intent, and what becomes of the four existing pages already half-addressing it. Volume tools understate GCC demand systematically, because the same question is asked in English, in Arabic and in transliterated Arabic, and often with a district name rather than the city. The terms with the best commercial fit routinely report as zero. The decision is what to publish against a number you cannot yet see, and that is a commercial judgement rather than a research one.
“Write a 2,000-word article on this topic that will rank.”
Genuinely good writing, and better than a fair amount of what currently ranks above you.
What the decision requiresRanking is decided less by the article than by what it is allowed to displace. If four of your own pages already touch the topic, a fifth divides the signal rather than adding to it, and the correct move is usually to merge three into one and redirect the remainder. Beyond that, a page earns citation through detail no one else holds: your own figures, your own photographs, a named author with a verifiable history. The prose is the cheapest part of the exercise, and it is the part that has become abundant.
What a year of doing it alone usually looks like.
- Publish more pages, more oftenA busier site that moves nothingMerging and redirecting pages that should not exist
- Redesign, URLs mapped from the sitemapRankings gone, redirects landing on the homepageRecovery costs more attention than the redesign did
- Trust green scores from a toolHealthy everywhere, empty reports where it mattersFinding the pages the tool could never reach
- Translate the site into ArabicArabic pages competing with the English onesRebuilding Arabic as a separate market entirely
- Let listings and profiles drift apartClean site, and still nothing resolvesReconciling every external mention to a single description
If you run it yourself, run it in this order.
- Before anything elseEstablish what is actually indexedPublish nothing until you know which pages can be seen
- Next, once onlySettle identity before contentContradiction is the most expensive problem, and the cheapest to remove
- Before writingMap questions and assign themEach question belongs to a single page, so pages stop competing
- Page by pageConsolidate before you createA site usually gets stronger, and faster, by shrinking first
- Once mappedPublish evidence, not coverageOnly what you can say better than anyone, under a name
- From then onMeasure retrieval apart from rankingAttribution needs a dated record of what was done, when
No one can guarantee AI answers — we improve the signals that earn them.
Six workstreams, drawn as one machine.
Six things must be true at the same time. When one of them quietly fails it suppresses the other five, which is why isolated fixes so rarely move anything.
What the PDF adds to this.
- —The six workstreams in full — what each one actually contains, not just what it is called.
- —One line item opened into twelve decisions — the part that decides the result.
- —The shape of a mandate — scope, cadence and what the work honestly asks of you.
- —The written argument behind every diagram on this page, in full.
Run it yourself.
Or hand it to someone whose name is on the result.
The tasks became cheap; who answers for the number at the end of the quarter did not change. We will tell you honestly if you are not a fit.
Request a Growth Audit