How the scores are actually computed.

The CITE score (GEO)

CITE scores a domain's visibility in AI-generated answers from 0 to 100 across four weighted dimensions: Citation (35%), Trust (25%), Identity (20%), and Eminence (20%). A structural failure, such as zero measured citations, caps the verdict at needs-fix regardless of the weighted total.

• Citation35%Whether AI answers to real buyer queries actually cite the domain, measured live across answer engines. Without measurement keys this dimension is estimated from on-page signals and flagged as estimated.
• Identity20%Whether machines can resolve who the brand is: structured data, entity clarity, canonical signals, consistent naming.
• Trust25%Signals engines lean on when choosing sources: crawlability by AI bots, server-side rendered content, freshness, contact and authorship signals.
• Eminence20%Third-party presence: whether the brand exists in the directories, reviews, and sources answer engines quote.

The SEO Health score

SEO Health scores classic search readiness from 0 to 100 across five weighted dimensions: Technical (25%), Content & E-E-A-T (25%), On-page (20%), Schema (15%), and Performance (15%). Every check is deterministic: the same input produces the same score, so month-over-month deltas are real movement, not noise.

• Technical25%Indexability fundamentals: robots, sitemap, canonical, server-rendered content, no stray noindex.
• On-page20%Title and meta description lengths, a single H1, image alt coverage, internal linking.
• Content & E-E-A-T25%Content depth, authorship identity, contact and trust signals, freshness, page citability.
• Schema15%JSON-LD presence, Organization or WebSite identity, type variety.
• Performance15%Page weight, script count, mobile viewport, HTTPS (heuristic; full Core Web Vitals run on paid plans).

Measured versus estimated

Two layers of the report are deterministic and always computed from fetched pages: technical access and page citability. The AI-visibility layer is a live measurement: real buyer queries are run against answer engines and each cell records whether the brand was cited. Live answers change run to run, so this layer is reported as a dated snapshot, never as a stable fact. When a cell cannot be measured, the report says unmeasured; it is never filled in with a guess.

There is a third case that most tools quietly score as a loss, and we do not. If your brand name is shared with something else, an answer engine sometimes resolves the query to the other entity: ask about a design studio whose name matches a developer tool, and the answer arrives full of software. We run an identity check on every mention, and when it says the answer is about a similarly named but different company, that cell is excluded in both directions. It is not a citation, and it is not a loss either, because the question was never about you. Counting it as a loss would mean reporting a defeat in a race you never entered, and it would quietly lower the CITE score of every brand with an ambiguous name. Excluded cells are labelled as such in the report, with their own count, separate from cells we simply could not reach.

Where this sits against Google's official guidance

Google publishes its own guide to optimizing for generative AI features, and its headline conclusion is unflattering to this whole product category: “Optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” We agree, and it is the reason BeCited ships two scores rather than one. SEO Health covers the fundamentals Google says still decide everything. CITE covers the one thing a rank tracker structurally cannot show you: an AI answer names a few brands and quotes a few third-party sources, and you are either in that set or you are not.

Google also describes how those answers are assembled: retrieval-augmented generation over its ranking system, plus “query fan-out,” where the model issues related queries of its own. That is the mechanism behind the query matrix in every report. We do not have access to Google's fan-out; we run a buyer-query set and record what came back.

What we are not

The same guide warns readers to “be wary of third-party tools that promise ranking success or claim to use ‘internal’ Google metrics,” and notes that “no third-party tool has access to our internal ranking or AI systems.” BeCited is a third-party tool. We have no access to any engine's internal ranking or AI systems, and nothing in a report is derived from one. Every number here comes from a page we fetched or an answer we prompted and recorded.

For Google specifically, the first-party ground truth is the Generative AI performance report in Search Console, and we would rather you read it than take our word for anything. BeCited exists because that report covers Google and your buyers also ask Perplexity, ChatGPT, Claude, Gemini, Grok, and DeepSeek. We measure across engines and show the sources; we do not replace first-party data where it exists.

What Google says you do not need

The guide has a mythbusting section, and every item in it is something a GEO tool could plausibly sell you. If our standard is that nothing unmeasured is presented as proof, the first place to apply it is our own feature list. Here is each claim and what we do about it.

• llms.txt and AI-specific files

  Google: “You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search.”

  Our llms.txt generator stays a free tool, and its own page says Google Search does not use the file. No CITE dimension awards points for having one.

• Chunking content for AI

  Google: “There's no requirement to break your content into tiny pieces for AI to better understand it.”

  The fix list never says shorten your paragraphs or add a Q&A block under every heading. When we raise a readability finding, the reason is a human reader, not a parser.

• A separate writing style for AI

  Google: “You don't need to write in a specific way just for generative AI search.”

  There is no GEO-formatted writing check to buy here. What we measure is not the shape of your prose; it is whether the answer named you.

• Structured data as a requirement

  Google: “Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add.”

  Schema stays where it belongs: 15% of SEO Health, for classic rich-result eligibility. In CITE, the Identity dimension asks whether your identity resolves at all, and does not mandate any particular markup.

• Collecting mentions

  Google: “Seeking inauthentic “mentions” across the web isn't as helpful as it might seem.”

  This is why the placement queue classifies surfaces. A directory profile or a real editorial surface produces work; a manufactured mention does not, and a competitor's own website produces none at all.

One honest caveat, because the distinction matters and is easy to blur. Every quote above is Google describing Google Search. BeCited measures seven engines, and the others have made no equivalent statement. Where an engine has not said what it uses, we report what we observed rather than assuming Google's answer applies to it.

Quotations on this page are from Google's AI optimization guide, used under the Creative Commons Attribution 4.0 License. BeCited is not affiliated with or endorsed by Google.

What the scores cannot claim

No score guarantees rankings, citations, or inclusion in any AI answer. Answer engines change their sources and behavior without notice. BeCited commits to measurement and to the work the fix list describes; it does not commit to platform behavior. Every recommendation in a report traces back to a measured gap on this page's terms.