Six signals AI assistants evaluate when identifying business recommendations
Across the major generative search interfaces, six characteristics emerge consistently as the signals that drive citation. The list is not exhaustive, but a firm that has not addressed these six is, in my work, materially less likely to be recommended. A framework for assessing where the firm stands.
Across the engagements I've run in the last twelve months, and across the test-query work I've done against ChatGPT, Claude, Perplexity, Microsoft Copilot, Google's AI Overviews and Gemini, six signals emerge consistently as the characteristics that drive a firm's inclusion in a generative search recommendation.
The list isn't derived from formal academic research; the underlying models are closed, and the precise weighting the platforms apply isn't disclosed. The signals are derived from observed behaviour across hundreds of test queries, and from triangulating against the firm characteristics that correlate, in my work, with citation.
The intent of the framework is to give a firm a structured way of assessing where it stands. A firm that scores well on five of six is materially more likely to be recommended than a firm that scores well on two.
Signal 1: declared identity
The most fundamental signal is whether the firm has a clear, machine-readable declaration of what it is. This is the foundational schema.org work: Organization or ProfessionalService markup, with a complete property set, on every page. (Covered in detail in the practitioner's guide.)
What the model checks for, in effect:
- Does the firm have a stable, declared name?
- Does the firm declare what it does, in plain language, in a way that matches the visible content of the homepage?
- Does the firm declare its jurisdiction and area served?
- Does the firm declare a principal whose reputation corroborates the firm's claim of expertise?
The absence of structured-data identity is the single most common reason a firm is invisible to AI search. The fix is well-bounded.
Signal 2: corroboration across sources
The model wants to verify the firm's claims against independent sources. The mechanism is the sameAs linkages from the firm's structured data to its presence on:
- LinkedIn (firm page and principal profiles)
- Companies House (or the jurisdiction's equivalent)
- Recognised professional body directories (ICAEW, RICS, Law Society, SRA, accredited body listings)
- Crunchbase or other firmographic databases
- Wikipedia, where applicable
- Industry publications and trade press listings
A firm that has a clean LinkedIn presence and is registered with Companies House but hasn't declared either in its sameAs is making the model do more work than necessary. The model may or may not bridge the gap; declaring it removes the uncertainty.
Signal 3: content depth and specificity
The model needs material to quote. A firm whose pages are all 400-word marketing summaries is harder to cite than a firm whose service pages average 1,500 words and contain specific, quotable claims.
What the model checks for, in effect:
- Are the firm's service pages substantive?
- Do the pages contain specific claims (numbers, dates, outcomes, methodology) the model can extract?
- Are there canonical-answer pages (FAQs, glossaries, explainer pieces) the model can use to answer specific queries?
- Is there a current blog or insights section with material the model can use to assess the firm's current thinking?
The cost of meeting this signal is where firms most often resist. The work isn't technical; it's editorial. The Authority package is the productised version of the sustained content investment this signal requires.
Signal 4: consistency
The model down-weights sources whose claims contradict each other or whose claims contradict other sources the model trusts. The internal-consistency check is straightforward:
- Does the firm's website say the same thing as its LinkedIn?
- Does the firm's website say the same thing as its third-party directory listings?
- Do the firm's pages say the same thing as each other?
- Are the principal's biographical claims (years of experience, credentials, prior firms) consistent across the firm's website and the principal's other professional listings?
The most common failure on this signal is the principal's LinkedIn profile being more detailed than the firm's website, with material the website omits or, worse, contradicts. The fix is to align the two.
Signal 5: recency
The model materially prefers recent material on most topical questions. The bias is particularly strong for:
- Regulatory or compliance topics
- Topics that involve technology or tools that are themselves moving quickly
- Market commentary
- Pricing and commercial structure
The implication for the firm is that a sustained publishing cadence is no longer optional. A firm whose last published piece is from two years ago is, on most queries, structurally disadvantaged against a competitor publishing monthly.
The bar isn't high. A firm that publishes a substantive piece every six to eight weeks, on the topics it wants to be associated with, will materially out-perform a firm publishing only sporadically.
Signal 6: explicit machine-readable summaries
This is the signal where the firm's llms.txt and its FAQ markup interact. The model is looking for a place where the firm has, in effect, written down the canonical answers to the questions a prospect would ask.
The firm that has both:
- A
llms.txtcovering the firm at the level of the organisation. (See Understanding llms.txt.) - FAQ markup covering the canonical questions on each major service page. (See FAQ schema.)
is materially more likely to be cited than a firm with only one or neither. The two signals are complementary; they cover different scales of summary.
How to use the framework
The framework is most useful as a structured audit. A firm can be scored, signal by signal, on a four-level scale (absent, partial, present, exemplary), and the resulting profile makes it immediately obvious where the priority work lies.
The audit framework piece sets out the full assessment criteria, with checklists for each signal. The Audit package is the productised version of that exercise.
What this means for the firm
Most SMBs I audit score well on one or two signals and materially under-perform on three or four. The Pareto position is to address the structural signals first (Signals 1, 2, 6) and then the content signals (Signals 3 and 5). Consistency (Signal 4) tends to resolve itself once the other work is done.
A firm that addresses all six is, in my observation, more likely than not to appear in the cited material for the queries it is positioned to answer. That's the underlying objective of the discipline.