After two years of practitioners running structured visibility audits across the major AI assistants, a recognisable pattern has emerged. Inclusion in AI-generated answers is not random and not purely a function of traditional SEO authority. Seven concrete signals explain most of it.
Marketing and PR teams that have spent the last two years trying to understand why their brand appears in some AI assistant answers and not others have, by now, ruled out the easy explanations. It is not purely a function of domain authority — high-DA brands are routinely absent from generated answers in their own categories. It is not a function of how often the brand is mentioned online — brands with substantial mention volume can still be invisible inside generated responses. It is not random — the same query asked across days returns recognisably similar sets of cited brands, and changes in the response set track to identifiable changes in the underlying information ecosystem.
What is left, after the easy explanations are ruled out, is a set of seven concrete signals that consistently differentiate brands that appear in AI-generated answers from those that don’t. The signals are not equally weighted across assistants, the weighting shifts over time, and the relative importance is category-dependent. But the seven are stable enough that practitioners working in this space have started using them as the operating framework for AI visibility work.
What follows is the working version of those seven signals, derived from observable patterns across the major AI assistants — ChatGPT, Gemini, Perplexity, Claude, Google AI Overviews — through the first half of 2026. Treat this as a practitioner’s framework, not a published research finding.
Signal 1: Structured, machine-readable description of the entity
The most reliable single predictor of brand inclusion in AI-generated answers is whether the brand has a clear, structured, machine-readable description of itself that the AI retrieval systems can extract and use. This includes: a well-maintained Wikipedia entry (or in non-English markets, the local equivalent), accurate Wikidata properties, schema.org structured data on the brand’s own website, and consistent NAP (name, address, phone) information across the major directory and social platforms.
The mechanism is straightforward. When an AI assistant generates an answer that requires it to describe an entity — a brand, a product, a company — the system pulls from structured sources first because they are higher confidence than free-form web text. A brand without these structured representations forces the assistant to assemble a description from less reliable sources, and the assistant may simply skip the brand if the assembled description is uncertain.
The work this implies is unglamorous and recognisable: build or improve the Wikipedia entry where the brand qualifies for one, get the Wikidata properties right, implement organisation and product schema on the website, and audit the directory consistency.
Signal 2: Citation-class third-party coverage
The second signal is whether the brand has been covered by sources the AI retrieval systems treat as authoritative. The list of citation-class sources varies by AI assistant and shifts over time, but it consistently includes major news outlets, established trade publications, government and academic sources, and a tier of specialised review sites that have built credibility in specific categories.
What counts as citation-class is something the practitioner has to learn through observation rather than from a published list. The practical method is to run sample queries in the assistants and look at what sources are being cited. Patterns emerge quickly: in a given category, the same six to fifteen sources tend to be cited repeatedly. Coverage in those sources moves a brand into the cited set; coverage in lower-tier sources tends not to.
The strategic implication is that PR work has shifted from being a brand-awareness activity to also being an AI-visibility activity. The placement that lands in a citation-class source is now doing double work — building awareness in its own right, and feeding the AI retrieval systems with a high-confidence source that includes the brand.
Signal 3: Topical authority on the specific question
The third signal is whether the brand has produced content that is recognisably authoritative on the specific topic the AI assistant is answering. The granularity here matters. A brand can be authoritative on a broad category without being recognised as authoritative on a specific sub-topic, and the AI systems are resolving at the sub-topic level.
The practical work is question research that goes deeper than category keywords. The team identifies the specific questions the AI assistants are getting asked in the category, then produces content that is a credible, sourced answer to those specific questions. The content needs to be substantive — short or templated content underperforms — and it needs to be discoverable to the retrieval systems through the standard surface area (a public, crawlable, well-structured page).
The frequency with which “we have a great blog” turns out not to translate into AI visibility usually comes down to this signal. The blog produces volume but does not authoritatively address the specific questions the assistants are being asked.
Signal 4: Cross-source consistency
The fourth signal is whether the description of the brand is consistent across the sources the AI retrieval system is consulting. When a brand is described slightly differently in five different sources — one says it’s a SaaS company, one says it’s a software vendor, one says it’s a digital platform, one says it’s a tool, one says it’s an agency — the assistant resolves the inconsistency by simplifying or by omitting. Both responses are bad for the brand.
The work is editorial consistency across the surface area of the brand’s public presence: website, social profiles, directory listings, press releases, partner pages. The descriptions should converge on a small number of clear, mutually-consistent framings, not diverge across the surface area.
Practitioners working with brands that have grown through acquisitions, brand refreshes, or pivots commonly find this signal underwater on day one of an AI visibility audit. The fix is mechanical but takes time.
Signal 5: Recency and freshness of content and mentions
The fifth signal is whether the brand has recent content and recent third-party mentions. AI assistants weight recency variably across categories — in fast-moving categories (technology, news, regulated finance) the weighting is heavy; in stable categories (industrial, infrastructure) the weighting is lighter — but recency is consistently a factor.
The practical implication is that AI visibility favours brands with steady content production cadence and steady earned media cadence over brands with bursty patterns. A brand that produced 50 pieces of content in 2023 and 0 in 2025 is less visible in mid-2026 than a brand that has produced 5 pieces per quarter steadily through the same window. The same applies to third-party mentions.
The work is the discipline of cadence, not heroics. A monthly content rhythm sustained over years beats a quarterly content sprint.
Signal 6: Sentiment and framing of available descriptions
The sixth signal is the sentiment and framing of the descriptions the AI retrieval system finds. When the available sources describe the brand neutrally or favourably, the brand makes it into the generated answer; when the available sources include substantial negative framing or controversy, the brand often gets omitted or the generated description softens.
This signal matters most for brands that have had a public incident — a controversy, a scandal, a high-profile customer complaint cycle, an unflattering investigative piece. The AI assistant’s response to the brand’s name is shaped by the residue of that incident in the retrievable sources, often more durably than the brand expects.
The work is reputation management, but reframed: the goal is not just to manage perception in human-readable channels but to ensure that the retrievable description of the brand across the AI-relevant sources is balanced, current, and accurate. This sometimes requires producing or sponsoring credible third-party content that reframes a contested narrative; it sometimes requires patient work to age out old content from the active retrieval window.
Signal 7: Distinctive, retrievable proof points
The seventh signal is whether the brand has distinctive proof points that are retrievable and usable by the AI assistant in framing an answer. Proof points are claims that differentiate the brand and are substantiated in a way the retrieval system can verify: specific customer outcomes, third-party rankings, awards, technical benchmarks, certifications, case study results, recognised partnerships.
The mechanism is that AI-generated answers about a category tend to be comparative — the answer describes multiple brands and what each does well. A brand with distinctive, retrievable proof points gets a positive framing inside that comparison; a brand without them either gets a generic framing or gets omitted entirely in favour of brands that have the proof points.
The work is producing and surfacing the proof points in a way the retrieval system can pick up. A case study that exists as a private PDF behind a form is invisible to the system; the same case study published as a public, structured, indexable page is retrievable. An award is invisible if it lives only on the partner’s site; it becomes retrievable when it’s also covered on the brand’s own page and in third-party sources.
How to operationalise this
The seven signals translate into a structured audit and a working programme for any marketing or PR team taking AI visibility seriously.
The audit runs the brand’s name and the brand’s category through a defined query corpus across the major AI assistants and captures the outputs. The team scores each captured output against the seven signals: which are present, which are weak, which are missing. The audit produces a prioritised list of gaps and the work required to close them.
The working programme runs the seven signals as parallel workstreams owned by the appropriate functions: structured data and Wikipedia work owned by the digital team, citation-class PR owned by the comms team, topical authority content owned by the content team, cross-source consistency owned by brand or marketing operations, content cadence owned by content production, sentiment management owned by comms or PR, distinctive proof points owned by the product or customer success team.
The monitoring layer runs the same audit at quarterly cadence to measure progress. The expected pattern is that the brand’s visibility inside AI answers improves steadily across two to four quarters as the signals come up, with the categories where the brand was weakest moving fastest.
The tooling category that has emerged to support this work is sometimes called AI visibility monitoring or generative engine optimisation analytics. Tools in this category — including UNmiss.com and a small set of competing platforms — run the persistent query corpus, track the outputs over time, and produce the reporting layer that lets the team see whether the signals work is producing results. The tools are not magic; they are the analytics layer that makes the seven-signal work measurable.
What this is not
This is not a checklist that guarantees inclusion in AI-generated answers. The AI retrieval systems are evolving, the weights are shifting, and brand inclusion is also competitive — being strong on the seven signals matters relative to how strong the competing brands in the category are on the same signals. A brand that has the seven signals at a “B” level in a category where competitors are at “C” will perform well; the same brand in a category where competitors are at “A” will struggle.
This is also not a substitute for traditional brand-building. The seven signals describe what AI retrieval systems use to determine inclusion in generated answers; they do not describe how the brand should think about its actual value to customers. The strongest performers on AI visibility are also the strongest performers on the more fundamental brand work, and the underlying brand strength is what makes the AI visibility work productive.
Final note for practitioners
The single most useful realisation when starting this work is that AI visibility is a discipline with concrete inputs and measurable outputs, not a mystery. The seven signals are not a complete model — they are the working approximation that has held up across two years of practitioner observation — but they are concrete enough to act on, prioritise, and report against. For marketing and PR teams that have been frustrated by the opacity of “why are we not in the AI answer?”, the seven signals are a much more useful starting point than the alternative of treating the question as unanswerable.
