Ferguson Media Group

Search & AI

SEO vs LLM optimisation: what is actually changing?

Most of what works for search still works. The differences are real but narrower than the current volume of advice would suggest.

Last reviewed

Search optimisation and LLM optimisation overlap substantially: both reward useful, well-structured, accurate, well-sourced content on a technically sound site. The genuine differences are in emphasis — answer-first structure, paragraphs that make sense in isolation, explicit entity naming and consistency of facts across a site all matter more when a system is summarising content than when it is ranking it.

What carries over unchanged

  • Genuinely useful content. Both systems are attempting to identify it; they differ in method, not in objective.
  • Technical accessibility. A page that cannot be crawled cannot be ranked or retrieved.
  • Semantic HTML. Headings, lists and tables communicate structure to everything that reads a page.
  • Server-rendered content. Anything that only exists after JavaScript runs is at risk of being missed by both.
  • Internal linking. Related resources connected in context help a reader, a crawler and a retrieval system alike.
  • Site speed and stability. Still the difference between a visitor staying and leaving.

What genuinely differs

Where the emphasis shifts between ranking and summarisation
ConsiderationConventional searchAI answer systems
Unit of competitionThe page, competing for a positionThe passage, competing to be the source of an answer
Where the answer belongsAnywhere on the page, provided the page is relevantNear the top, stated directly, before the explanation
ContextThe surrounding page supplies itThe paragraph may be read alone and must carry its own context
Entity referencesPronouns are generally resolved from contextExplicit names are considerably safer
Consistency across a siteMatters modestlyMatters a great deal — contradictions undermine the whole source
The measurable outcomeClicks and positionsCitation and mention, both of which are hard to observe

The measurement problem

Conventional search offers imperfect but real feedback: positions, impressions, clicks. AI answer systems offer very little. A page can be summarised for thousands of people with no visit, no log entry and no record of any kind.

That has two practical consequences. Anyone claiming to measure AI visibility precisely is overstating what is observable. And the sensible response is to focus on the things that can be controlled — clarity, structure, accuracy, sourcing — rather than on optimising towards a metric that does not exist yet.

What to actually do differently

  1. Move the answer to the top

    Open each page with a direct, complete answer to the question it addresses, then explain. This costs nothing in readability and frequently improves it.

  2. Write paragraphs that survive extraction

    Name the subject. "Domain brokerage is..." rather than "It is...". Read a paragraph in isolation and see whether it still makes sense.

  3. Put facts where facts belong

    Specifications in tables, definitions as definitions, figures with units and dates. Prose is for argument; data should be structured.

  4. State who says so

    Named sources, visible dates, stated methodology. Verifiability is difficult to fake at scale, which is precisely why it counts.

  5. Audit for contradictions

    The same fact stated differently on two pages weakens both. This is the single most overlooked item on most sites.

What has not changed at all

Nobody can guarantee a ranking, and nobody can guarantee a citation. Both systems are controlled by organisations that do not publish how they select, and both change frequently.

The durable position is the same one it has always been: publish something genuinely worth referencing, make it easy to understand, and be honest about what you know. That approach was correct before any of this and remains correct now — which is the argument made in full on the LLM and AI search optimisation page.

Questions

Should I stop investing in SEO?

No. Most of what constitutes good search work is what makes a site legible to AI systems as well. What is worth stopping is the production of pages that exist only because a keyword had volume, since those were the pages most exposed to being answered directly.

Do I need a separate AI search strategy?

Not a separate strategy, but a deliberate review. Most sites need a handful of structural changes — answer-first openings, clearer entity naming, consistent facts, better sourcing — rather than a parallel programme of work with its own budget.