The Generative Engine Optimisation Checklist for 2026

Two years ago the objective was a blue link in position one. It still matters, but for a growing share of queries the assistant answers in-line and the click never happens. The winning position in those cases is not a rank, it is a citation: being the source the model quotes and links when it composes the answer. That is what generative engine optimisation is for, and it is now a distinct discipline sitting alongside conventional SEO rather than replacing it.

This is the checklist we work through on client sites. It is ordered by leverage, so if you only do the first three sections you will still see movement. The broader service context is on our AI SEO and search visibility page, and the strategic argument in generative engine optimisation: getting cited in AI answers.

1. Make your entity unambiguous

Models resolve queries to entities before they resolve them to documents. If a system cannot confidently determine what your organisation is, what it does and where it operates, it will not cite you, because citing an ambiguous source is a risk to answer quality. Entity clarity is therefore the highest-leverage item on this list and the one most sites fail.

What to fix

• One canonical organisation name, used identically everywhere including the footer, schema, social profiles and directory listings
• A single Organization schema node with a stable @id, referenced by every other schema node on the site
• Consistent NAP details: name, address and phone identical across the site, Google Business Profile and third-party listings
• An about page that states plainly what the organisation does, for whom, and where, without marketing abstraction
• sameAs links from your Organization schema to the profiles that corroborate the entity

The corroboration point is what most teams miss. A model gains confidence when independent sources agree. One page asserting a fact is a claim. Five consistent sources asserting the same fact is a fact worth repeating in an answer.

2. Ship structured data that a machine can trust

Structured data does not make content rank, it makes content legible. For generative retrieval that legibility matters more than it did for classic search, because the model is extracting specific claims rather than assessing overall relevance.

The nodes worth having

1. Organization on every page, with logo, contactPoint, address and areaServed
2. Service or ProfessionalService on each service page, with an offer catalogue rather than a bare name
3. BlogPosting on every article, with datePublished, dateModified, author and articleSection
4. FAQPage on pages that genuinely answer discrete questions, matching visible on-page content exactly
5. BreadcrumbList on every page below the root, so hierarchy is explicit rather than inferred
6. AggregateRating and Review where you hold genuine, verifiable reviews

Two rules govern all of it. Schema must match what a visitor sees on the page, because mismatch is treated as a trust signal in the wrong direction. And review or rating markup must reflect real, attributable reviews, since fabricated ratings are both a policy violation and, in most jurisdictions, a consumer protection issue.

3. Write answer-shaped content

Content built to be extracted looks different from content built to hold attention. It front-loads the answer, keeps each claim contained in a single passage, and uses headings that state a question or a conclusion rather than a theme. Models retrieve passages, not pages, so a passage that only makes sense after reading two sections above it is a passage that will not be quoted.

Practical structure rules

• Answer the title's implicit question within the first hundred words
• One idea per paragraph, self-contained enough to stand alone if lifted
• Descriptive H2 and H3 headings that would work as a standalone question or claim
• Lists and tables for comparisons, steps and specifications, since these extract cleanly
• Specific figures, dates and named methods rather than qualitative hedging

Specificity deserves emphasis. "Significant improvement" is unquotable. "Rep productivity rose 79% in three months" is quotable, verifiable and memorable, which is exactly the profile of a passage that gets cited.

4. Build internal links as an authority graph

Internal linking is not navigation. It is how you tell a retrieval system which of your pages is the definitive source for a given concept. Every time an article links the phrase "AI CRM development" to the same destination, that page's association with the phrase strengthens. Inconsistent anchors dilute the signal across several pages and none of them wins.

The pattern that works

1. Nominate one canonical page per core key phrase, usually a service or industry landing page
2. Always link that phrase to that page, with the key phrase itself as the anchor text
3. Link every new article to at least two landing pages and two related articles, in context rather than in a footer block
4. Link back from landing pages to the strongest supporting articles, so authority flows both ways
5. Keep every page reachable within three clicks of the home page, and maintain an HTML sitemap as a backstop

That last point matters more for generative retrieval than for classic crawling, because a page that is hard to reach is a page that is under-represented in whatever index the model consults.

5. Cover the long tail at a volume manual production cannot reach

Head terms are contested and increasingly answered without a click. The queries that still convert are long, specific and intent-heavy: a service, a qualifier, a location, a constraint. There are thousands of them, each with modest volume, and the economics of hand-writing a page for each has never worked.

Programmatic generation solves the volume problem, but only under conditions. Each page must be gated on real data, a real offer and real intent. If you cannot say something genuinely specific about a combination, do not publish the page for it, because thin templated pages harm the entity clarity you built in section one. Done properly, this is the single largest source of incremental qualified traffic available to most B2B sites, and it is the approach behind our data platforms and enrichment work.

6. Keep the technical foundations current

• Server-rendered HTML for all indexable content, since not every retrieval agent executes JavaScript
• Self-referencing canonical tags on every page, including paginated variants
• An accurate XML sitemap plus an HTML sitemap linked from the footer
• robots.txt that permits the AI crawlers you want citations from, and blocks only what you genuinely need to withhold
• Fast first-byte times, because retrieval systems operate under timeouts that browsers do not
• dateModified kept honest, since freshness is weighted heavily in generated answers

The robots.txt decision is a genuine choice with a trade-off. Blocking AI crawlers protects content from training use and also removes you from the answers those systems produce. For most service businesses the citation is worth more than the protection, but it should be a decision rather than a default.

7. Measure both winners' circles

Rank tracking alone will understate performance in this environment, because you can gain citations while losing impressions on head terms. Track four things in parallel.

1. Classic rank and impressions from Search Console, segmented by head and long-tail
2. Citation presence: run your priority queries through the major assistants monthly and record whether you are cited
3. Referral traffic from assistant domains, which is small in volume and unusually high in intent
4. Page-level conversion, since a page that earns citations and no enquiries is a content problem rather than a visibility one

The monthly citation check is manual for now and worth the hour. It is the only direct read on whether the first six sections are working, and it usually reveals which specific passages are being lifted, which tells you what to write more of.

Where to start

If the list feels long, do sections one and two first. Entity clarity and structured data are one-off engineering tasks with compounding returns, and most sites gain measurable citation presence from those alone. Sections three to five are ongoing content discipline, and section seven tells you whether any of it is landing.

If you want an audit against this checklist on your own domain, get in touch and we will run it, including the citation baseline, before recommending anything.