A language model does not rank your page.
It decides whether to name you at all.

It starts with your real buying queries, run live against each engine in scope, and records four things per result: whether your brand was named, whether it was linked, who was cited instead, and which page of theirs the model pulled from.

Every gap is then traced back to a cause in one of six layers: citation presence, entity and Knowledge Graph coverage, structured data, semantic coverage, authority signals and crawler access. The raw answers are attached to the report with timestamps, so any finding can be re-run and checked independently.

Two things are happening at once. The model has learned associations during training, which is why a well-established brand gets mentioned even without a live search. Alongside that, most assistants retrieve pages while they answer, and those retrieved sources are what usually get cited with a link.

A source tends to earn that citation when the answer to the question is present and self-contained on the page, the page states who is making the claim, and something outside your own website supports it. Content that reads well to a human but only makes sense in the context of the surrounding paragraphs is the most common thing we find sitting between a brand and a citation.

Yes, though the mechanism is worth being precise about. No engine publishes a rule saying valid schema earns a citation. What schema does is remove ambiguity: it states in machine-readable form what a page is, who published it, which organisation the claims belong to, and how that organisation connects to its profiles elsewhere.

For Google surfaces the benefit is most direct, because Gemini, AI Overviews and AI Mode are grounded in Google's own index and entity understanding. Across the other engines, schema mainly earns its place by making the entity behind a claim unmistakable. We implement Organization, Service, Product, LocalBusiness, FAQPage, Article and Author types, validate them, and confirm they agree with the visible page.

An entity is a thing a machine can hold onto: your company, your founder, your service, your location. Entity optimization means describing each of those consistently enough, in enough places, that a system treats them as one known thing rather than several loose mentions of a similar name.

The Knowledge Graph is Google's structured record of those entities, and a brand with a clear presence there is easier for any retrieval system to recognise and describe accurately. The practical work is unglamorous and effective: one canonical description, matching details across every profile and directory, explicit sameAs links in your markup, founders and credentials tied to the organisation, and disambiguation from any similarly named company in another category.

It is a decision worth making deliberately rather than by default. Blocking GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended or CCBot reduces the chance of being cited by the assistant that depends on it, so the choice is about which trade-off suits your business. A common approach is to allow the agents that power live search and citation while being more selective about the ones used purely for training.

Something we check on nearly every audit: accidental blocks. Firewall rules, CDN bot management and inherited robots.txt lines block AI crawlers far more often than anyone intends. On llms.txt, it is a proposed convention rather than an adopted standard, and support for it is limited today. It costs almost nothing to publish, and we will add it where it is useful while being clear that it is not the thing that earns citations.

They overlap heavily, and any agency claiming the two are unrelated is overselling the novelty. Crawlability, rendering, site structure and topical authority feed both, and a site with technical problems will struggle in either.

What is genuinely additional is the measurement layer and the entity work. Rank tracking does not tell you whether ChatGPT names you, so the queries have to be run against each engine directly and recorded. And because a citation is an attribution decision, entity clarity and off-site corroboration carry more weight than they do for a ranking position. Our own view is that this belongs alongside search work rather than in a separate silo, which is why the reporting sits on one timeline.

It varies by engine and by the type of fix, so it is worth setting expectations by mechanism rather than by a single number. Engines that retrieve live pages while answering, Perplexity most visibly, can reflect a change within days of the page being recrawled. Surfaces grounded in a search index move on that index's timeline, so weeks is more realistic.

Associations learned during training move slowest of all, because they only update when a model does. That is the honest reason entity clarity and off-site corroboration sit at the top of most fix lists: they are the signals that compound across both the retrieval layer and the next model release.

The implementation of the fix list, and the tracking that shows what it did. In practice that means schema built into your templates, content restructured so key claims survive extraction, entity and profile consistency work across the sources that describe you, off-site corroboration where a claim currently stands alone, and crawler access configured deliberately.

Alongside it, the original query set is re-run on a schedule and reported against the baseline, so movement is measured rather than assumed. Everything is documented as it ships, and the configuration is handed over so your team keeps the visibility if the engagement ends.

The audit is genuinely useful on its own. It gives you a dated baseline, the reason behind each gap, and a fix list your own developers can act on. Plenty of value sits in that document alone, and it is yours to keep either way.

What a single snapshot cannot do is tell you whether an answer changed because of your work or because a model was updated. Answers shift as engines change how they retrieve and rank sources, so re-running the same query set on a schedule is what turns the baseline into a trend. Most clients start with the audit, act on the top of the list, then add tracking once they can see what moved.

Yes, and it is a common setup. In-house teams usually own content, keyword strategy and stakeholder relationships, and want engineering capacity for the entity, schema and structure layer rather than another strategy document.

We work to your ticketing process, document every change, and hand over the query set, the tracking configuration and the monitoring so your team keeps the visibility after the engagement.