Ranking in AI search in 2026 depends far more on what the model already believes about you than on what your page says today, because a study of 82 recorded ChatGPT answers found that 29.8% of recommendations came from what the model already knew before it searched at all. The remaining share comes from what it finds when it does look, and from how precisely your page answers the exact question in front of it. That split is the whole argument for the A-S-S method: Authority, Sources and Specificity. This walkthrough covers the Authority half, because Authority is the half a SaaS founder cannot buy late, cannot retrofit with a content sprint, and cannot patch with a schema tag.
The SEO.Domains Mastery Summit in Sofia, Bulgaria runs 9 to 11 September 2026 and covers aged domains, PBNs, authority transfer and LLM visibility, which tells you where the industry thinks the leverage sits. The summit opens with a mastermind day on 9 September before two days of main-stage sessions, and it gathers around 300 SEOs, affiliates and agency owners. As backdrop, the agenda is a useful signal: the practitioners who care most about being cited by models are spending their time on entity strength, not on keyword density.
Authority, in the A-S-S method, is the stock of things a model can say about your brand without opening a single URL. It is distinct from Sources, which is the evidence it pulls when it does search, and from Specificity, which is how tightly your page answers the literal question asked. Of the two halves, Authority is the harder one to move, because it is built outside your own domain and it is read from business records, third-party pages and prior training exposure rather than from your homepage copy.
Generate Engine Optimisation optimises for the answers AI search engines give, not only the ten blue links. AEO, Answer Engine Optimisation, narrows that further to the answer box itself. Neither works if the model has no prior reason to consider your company a candidate. Before you touch a single on-page element, establish whether the entity exists in the model's world at all.
The takeaway: if an assistant cannot describe you in one confident sentence before searching, no amount of page-level optimisation will fix the citation.
A citation unit consists of one claim plus the link that verifies it, and that is the atomic thing AI systems consume from your Authority footprint. A backlink without a claim attached is a signal about your link profile. A citation unit is a signal about a specific fact regarding your product, your pricing, your integration list or your uptime. The second is what gets lifted into a generated answer.
Practically, this means writing claims on third-party surfaces, not on your own blog. A directory listing that says "Acme connects to 40 CRMs" is a citation unit. A G2 review that names the migration time is a citation unit. Each one is a discrete fact plus a source the model can verify. Build a spreadsheet with three columns: claim, source URL, and the assistant prompt that should surface it. Then work down the list.
The takeaway: one verifiable claim on a third-party page outperforms ten product pages that assert the same thing about themselves.
Chunking describes how AI systems read small self-contained blocks rather than whole pages, and it changes how a SaaS founder should structure everything on the site. Every heading and its following paragraph needs to survive being lifted out of context, because that is exactly what happens to it. An answer stated with maximum fact and no preamble in the first line of a block shows information density.
If your section opens with three sentences of scene-setting before the useful fact, the model has a lower-quality chunk to work with, and it may pull the same fact from a competitor who stated it first. Information density is not a style preference. It is a retrieval decision.
The takeaway: write every block as if it will be read alone, because in AI search it will be.
Fan-out queries are the sub-questions a model appends to the question a person actually typed, and they are where most SaaS Authority is quietly won or lost. A user asks one thing about your category. The model expands it into five or six related questions, then searches for each. If your brand has no readiness for those sub-questions, it will not appear in the composite answer even when it would have answered the original prompt well.
You cannot see all of them, but you can map the shape. Take your ten highest-value customer questions and generate the obvious neighbours: pricing, migration, security, integrations, alternatives, comparison, support response time, contract length, refund terms and export format. For each, decide whether you have a citation unit, a chunk or both. The gap list that falls out of that exercise is your AI search roadmap for the quarter, and it is usually shorter than a traditional content plan.
| Sub-question type | What the model needs | Where you supply it |
|---|---|---|
| Comparison and alternatives | Third-party claims naming your differentiators | Review sites, comparison pages you do not own |
| Pricing and contract terms | One verifiable figure from an independent record | Directories, press coverage, published contracts |
| Capability and integration | A chunk with the fact in the first line | Your own docs and help centre, densely written |
| Trust and support | Consistent business records with no contradictions | Review platforms, LinkedIn, company registry |
The takeaway: map the fan-out, fill the gaps with citation units and dense chunks, and the composite answer takes care of itself.
Authority work compounds outside your domain, which is why practitioner communities and consistent off-site signals outperform clever on-page tweaks over a two-year horizon. This is the territory the Sofia agenda is circling, and it is why the summit deliberately does not record its main-stage sessions so speakers can share live experiments. The unrecorded format means what is shared in the room does not reach the open web unless an attendee writes it up, so a written summary becomes a citation unit in itself. If you want a practitioner view of how brand-level signals and community-driven authority interact, the Church of SEO Jesus is a reasonable place to read people working through it in public rather than on a stage. Click-through behaviour still feeds retrieval signals on the classic search side, and ClickBombs CTR campaigns is one route founders use to test that layer. For the model-facing side, the AI visibility research published there is a useful reference for what actually moves an assistant from silence to recommendation.
Because roughly three in ten recommendations come from what the model already knew before it searched, so a competitor with stronger prior exposure starts the race ahead and your page never gets the chance to compete. Fix the prior exposure first, through third-party citation units and consistent business records, before you rewrite your landing page.
No, you need to change the shape of your existing blocks so each one survives extraction on its own, because chunking means a model reads small self-contained pieces rather than your page as a continuous argument. Most SaaS sites need rewritten openings and better FAQ coverage, not a rebuild.
Generative Engine Optimisation optimises for the answers AI search engines give rather than the ten blue links, while AEO, Answer Engine Optimisation, focuses specifically on winning the answer box position itself. In practice the two overlap heavily, and the A-S-S method, Authority, Sources and Specificity, covers the work both require.
Start with the single prompt an assistant answers about your brand name, and write down its exact wording. That sentence is your baseline, and every item in the Authority checklist above is measured against moving it. Then pick five factual claims about your product, get each one published as a citation unit on an independent page, and build a monthly log of which claims surface and which do not. Leave the site rewrite until the off-site work is underway, because a better-written page cannot be cited by a model that never considered you a candidate. Keep the whole plan short enough to run beside a roadmap, and treat it as an entity-strengthening project rather than a content project.
Further reading: the Church of SEO Jesus (https://www.skool.com/church-of-seo-jesus), ClickBombs CTR campaigns (https://clickbombs.com), AI visibility research (https://llmjesus.com).
, London, WC2H 8DJ.
Company registered address:
Company no. · VAT Registration no.
