How is the SIGNALS score composed?

A SIGNALS score is composed as a weighted average of seven dimension scores, each marked out of ten, with the weights set by measured effect size rather than by preference. Alignment carries the most weight because it is the only dimension with a documented causal effect independent of domain authority.

The bars below are worth reading as exchange rates rather than as a ranking. On the weights set out in the SIGNALS methodology, a dimension at 35% is worth seven times a dimension at 5%, so a page that gains two points of Alignment has done more for its score than a page that gains a full ten points of Newness. That is the practical use of a weighted model: it tells you not just what is wrong but what each fix is worth, which is the question a team with one afternoon actually has.

Two of the seven behave differently from the rest and it is worth knowing which before you read the bars. Alignment cannot be assessed from the page in isolation, because it compares the page against what buyers type, and Substantiation cannot be changed from your own site at all. The other five are decided entirely by the words on the page in front of you.

What is a search signal framework, and is SIGNALS one?

A search signal framework is a named set of measurable properties that a search engine or an answer engine is believed to read, arranged so you can tell which one to fix first. Three things separate a framework from a checklist: the signals are named and defined, they carry a weight or an order, and the model says what to do when two of them disagree. SIGNALS is one, built as a weighted model of seven page-level signals.

The distinction that matters when you are comparing frameworks is how the signals combine. A weighted model, which is what SIGNALS is, gives every signal a share of one score, so a strong signal can offset a weak one and the instruction it produces is "fix the heaviest thing you are failing." A gated model runs the signals in sequence and treats the weakest one as a ceiling on the rest, so the instruction it produces is "fix this first, nothing after it will pay until you do." The Search Signal Framework published by Sunil Pratap Singh is the gated kind, organising visibility into four cascading signals: Presence, Trust, Relevance and Momentum. We have not independently tested that model, and the description here is of its published structure as of September 2026.

A comparison of model designs, not of vendors. Both designs are in use and they answer different questions.

In practice the two are complements rather than rivals. If your pages are not in the index at all, a weighted score is the wrong instrument, because it will hand you a list of content fixes that cannot pay until retrieval works. Once retrieval works and everything is merely imperfect, a weighted model is the better read, because the question has changed from "what is broken" to "which of these is worth doing this quarter." SIGNALS is built for the second question, which is why its output is a number with a breakdown rather than a single verdict.

Is a search signal framework the same as a signals framework in enterprise search?

No, a search signal framework of the kind this page describes and a signals framework in enterprise search are two different things sharing one name. In enterprise search it means a pipeline for behavioural data inside a search product you own. Here it means a weighted model of document properties that somebody else's retriever reads. People arrive at both from the same query, so it is worth settling which one you came for before reading any further.

Lucidworks publishes the clearest example of the first kind. User interactions are captured by a signals beacon or a platform signals API and sent to a signals store, where an aggregator turns the raw events into summaries, from simple sums to statistical functions. A query pipeline stage then runs a secondary query against the aggregated collection and converts those summaries into boost weights, so the items people engaged with rank higher for whoever searches next. Raw and aggregated signals sit in separate collections, and the aggregated documents have to be relatable to documents in the primary collection or the boosting cannot work at all (Lucidworks documentation, platform signals overview and the boost with signals stage, read 12 October 2026).

Two things called a signals framework, and what each one can actually change

	Behavioural signals framework, enterprise or commerce search	Citation signal framework, answer engines
What a signal is	An event: a click, a view, a cart addition, recorded against a document in your index	A property of the document itself: whether it states the answer, cites a source, carries a date
Where the data comes from	Your own users, through a beacon you install	The page as published. There is no event stream to collect
Who owns the index	You do, which is why the loop closes in days	The engine does, which is why nothing you measure is the ranking itself
What moves when a signal moves	The ordering of results inside your own search box	Whether an external engine quotes the page when somebody asks the question
Unit of measurement	A document in your collection, scored by engagement	A page on the open web, scored by its readable properties

Nothing crosses from the first column to the second, and the reason is worth stating because it is the same reason the click half of Google's own signals does not transfer. Engagement data in a Lucidworks-style framework is first-party and lives inside your index, so no external retriever can read it even in principle. The section below on Google's described signals works through the sharper version of that point, where Body transfers to AI citation, Anchors transfer partly, and Clicks do not transfer at all.

If you came here for the enterprise search meaning, the documentation for your own platform is the thing to read and this page will not help you. If you came here because an answer engine is not quoting your pages, the rest of this page is about that, and the two problems do not share a fix.

Which search signal frameworks are there, and how do you choose between them?

Seven named frameworks are in circulation for this problem, and they are not competing answers to one question. They measure different units, output different things, and are useful at different moments. Choosing between them starts with what you are about to decide rather than with which one sounds most complete.

Descriptions are of each framework's published structure, read September 2026, except GEO-16, read 16 October 2026. We have independently tested only our own.

E-E-A-T is the one most often mistaken for a scoring model. It is a set of criteria in Google's Search Quality Rater Guidelines, standing for Experience, Expertise, Authoritativeness and Trust, with Trust named as the most important member of the four and Experience added in December 2022 as first-hand evidence that the author actually did the thing (Google, Search Quality Rater Guidelines: An Overview). Raters do not set rankings, so E-E-A-T is a description of the target rather than a measurement of your distance from it. That makes it the right thing to read and the wrong thing to report against.

The GEO taxonomy sits at the other end. It is one of only two entries with measured citation data behind it, and the only one that measured what happens when a page is changed: adding relevant statistics, direct quotations and cited authoritative sources raised a source's visibility in generated answers by up to 40% on the paper's benchmark, while keyword stuffing and stylistic padding did not (GEO: Generative Engine Optimization, Aggarwal et al., KDD 2024). What it is not is a diagnostic. It tells you which kinds of edit are worth making in general, and nothing about which one your page is missing, which is the gap a page-level framework fills.

Which search signal frameworks have been tested against real citation data?

Two of the seven frameworks have been tested against real citation data, and only one of those is a scoring model you can run on a page. The other five publish a structure and an argument for it. That is not a disqualification, because a framework is a way of organising attention before it is a measurement, but it decides what you can honestly report to somebody who asks why these signals and not others.

GEO-16 is the tested scoring model. Kumar and Palkhouski put 70 product-intent prompts to Brave Summary, Google AI Overviews and Perplexity, collected 1,702 citations from the answers, and audited the 1,100 unique URLs behind them against sixteen pillars. Pages scoring 0.70 or better on their composite with at least twelve pillars hit were cited across engines 78% of the time, and overall page quality carried an odds ratio of 4.2 for citation in their logistic models, with a 95% confidence interval of 3.1 to 5.7 (Kumar and Palkhouski, "AI Answer Engine Citation Behavior: An Empirical Analysis of the GEO16 Framework", arXiv preprint, September 2025, confirmed from two independent readings on 16 October 2026 because this environment could not open the paper directly).

The same study is the strongest published argument for keeping one series per engine instead of one pooled number. The mean composite score of the pages each engine chose to cite was 0.727 for Brave, 0.687 for Google AI Overviews and 0.300 for Perplexity (same source and date). Perplexity cited pages that scored under half what Brave's did, so the bar is a property of the engine as much as of the page, and a page failing one engine's apparent standard can still be cited by another.

Where GEO-16 and this framework disagree is on which properties matter most, and the disagreement is worth stating rather than smoothing over. The study found that its strongest associations with citation were metadata and freshness, semantic HTML and structured data, all technical properties of the document (same source and date as above). SIGNALS puts 35% on Alignment, which is editorial, and 5% on Newness. Two explanations fit. The corpus was English-language B2B SaaS, a field where the editorial baseline is already high and the technical work is what separates one page from the next. And its pillars are counted as hits rather than weighted, so a property almost every candidate page has registers differently from how it would in a weighted model. We have not resolved which explanation is right, and a framework page that quietly dropped the inconvenient result would be worth less than one that names it.

The gap that remains, this framework included, is a validation of the composite score itself. The SIGNALS weights come from measured effect sizes in the studies cited against each dimension below, which is a narrower claim than "this total score predicts citation at this rate on this corpus." GEO-16 is the only one of the seven to have published that second kind of claim, and it published it on one corpus, in one vertical, with its authors calling the study observational, so it shows association rather than cause. If you are choosing a framework to report against, that is the fact to weigh, and it argues for treating all seven as instruments rather than as verdicts.

Which frameworks measure a brand rather than a page?

Three of the seven frameworks measure a brand rather than a page, and they are the ones to reach for when the problem is that engines describe you wrongly or not at all. SIGNALS cannot help with that, because it grades a document and the fault is not in any one document.

The Search Signal Framework published by Sunil Pratap Singh is the most explicit about the arithmetic. It scores each of its four signals out of 25 and treats the lowest as the ceiling, so a brand at 22 out of 25 on Trust and 8 out of 25 on Presence is capped by Presence and told to fix that first. Its published scope maps the 200 ranking factors in Backlinko's list across the four stages and adds 28 signals specific to generative engines, which is where the AI half of it comes from (Sunil Pratap Singh, The Search Signal Framework, read September 2026).

Rank4AI's Five Signal Model asks a narrower question, and a useful one: whether an engine can describe your business at all. Its five layers are Identity Clarity, whether the business can be described in one consistent sentence; Subject Authority, whether the engine associates you with expertise in something specific rather than with many things equally; Meaning Architecture, whether the site structure can be read and interpreted; Ecosystem Validation, whether independent sources confirm what the site claims; and Signal Consistency, whether the description has held still across the platforms an engine draws from (Rank4AI, The Five Signal Model, read September 2026). The failure it describes is one we see constantly in assessments, where a company is indexed, retrievable and still absent from answers because three of its own properties define it three different ways.

SIGNAL, published by Rajdeep Chauhan as a reputation framework, belongs in the census for a different reason. Its six components are Search, Intelligence, Governance, Narrative, Authority and Leverage, and only the first overlaps with what the other six frameworks here measure (Rajdeep Chauhan, SIGNAL Framework, read September 2026). It is listed because the name collides and a buyer comparing frameworks will meet it, not because it competes on the same question. If your concern is what an engine says about you rather than whether it cites you, that is the reputation problem and it is measured differently.

The honest summary is that you probably want two of them at once. Read E-E-A-T to know what you are aiming at, take the effect sizes from the research so the weights are not invented, then use a page-level or a brand-level model depending on whether your problem is "which of these fixes first" or "why is nothing working at all." SIGNALS answers the first question. It is the wrong instrument for the second, and we would rather say so than sell it as both.

Which search signals does Google actually use, and how do they map onto AI citation?

The search signals Google's own engineers describe are fewer than most framework pages imply, and only one of them maps cleanly onto AI citation. In testimony and a deposition released through the United States antitrust case, Google search engineer Hyung-Jin Kim described three core relevance signals, abbreviated ABC: Anchors, meaning the pages that link to a document, Body, meaning the query's terms as they appear in the document itself, and Clicks, meaning what users did after they saw the result. Those three combine into topicality, the query-dependent base score, which is then adjusted by a site-wide quality score and by click-based re-ranking that draws on about 13 months of data (Search Engine Land, The ABCs of Google ranking signals, and Search Engine Journal's report of the same deposition, both checked 7 October 2026).

That matters here because a buyer searching for a search signal framework is often really asking whether the signals they already know carry over. Mostly they do not. The table below maps each documented signal onto whether it transfers to answer engines and which SIGNALS dimension, if any, carries it.

Signal descriptions are from the antitrust disclosures cited above. The transfer column is our reading, not Google's.

Two practical conclusions follow. The first is that the signal most SEO work is aimed at, the click, has no equivalent in an answer engine, so a framework built on engagement metrics will not tell you why you are uncited. The second is that the one signal which transfers outright is the dullest of the three: whether the words a buyer uses appear in the document. That is why Alignment carries 35% here rather than being a tie-breaker, and it is also why the work is editorial rather than technical. If you would rather have that editing done than do it, we describe the engagement shape on our page about AI citation optimization services.

What do Structure and Intent measure?

Structure measures whether a page can be broken into extractable answers, and Intent measures whether those answers are the ones a buyer actually asked for. They are the first two because they act earliest: an extractor that cannot find the boundaries of an answer never gets to judge it, and an answer to a question nobody asked is not retrieved however well it is written. Each entry below gives the weight, the study the weight came from, and the checks that make up the dimension score.

What do Grounding and Newness measure?

Grounding measures whether the claims on a page carry a source in the paragraph that makes them, and Newness measures whether the page shows it has been maintained. They belong together because both are about how much a model can trust a sentence enough to repeat it. An unattributed figure is something the generator has to hedge, and a page with no date and no recent reference is something it has no reason to prefer over a fresher one.

Why is Alignment the heaviest signal in the framework?

Alignment is the heaviest signal because it is the only one that survived when the underlying study controlled for the domain a page sits on. Every other content property that looks predictive in a raw correlation turns out to be mostly a description of which sites already have authority. Alignment measures whether the page speaks in the vocabulary buyers actually type, and it is the one dimension that cannot be assessed from the page alone, because the comparison is against demand data rather than against the text.

What do Language and Substantiation measure?

Language measures whether a sentence can be quoted without the paragraph around it, and Substantiation measures what other domains say about you. The pair is worth reading together because they are the two most often misjudged in opposite directions. Language is treated as a style preference when it is a retrieval property, since a sentence that opens with "this" or "they" cannot be lifted into an answer on its own. Substantiation is treated as the whole problem when it carries 15% on the weights traced in the SIGNALS methodology, and it is the only dimension you cannot change by editing your own site.

Which signals can you change by editing a page, and which take months?

Four of the seven signals move on the day you edit the page: Structure, Grounding, Language and Newness. Intent moves when you add the sections you were missing. Alignment moves once you know what your buyers actually type, which is research before it is writing. Substantiation is the only dimension you cannot change on your own site at all, because it is a measure of what other domains say about you.

The weights make that split worth looking at as arithmetic, adding up the figures in the score composition above and in the SIGNALS methodology. Six of the seven signals, 85% of the score, are decided by what is on your own pages. The four that move with a single editing pass carry 40% between them. Substantiation, the one that depends on other people, carries 15%. A team that believes AI visibility is mostly an off-site problem has the ratio close to backwards.

Weights are the SIGNALS weights shown at the top of this page. The time column is our own estimate from running the work, not a measured figure.

The order in that table is the order of the weights, which is not the order most teams work in. Grounding, Structure, Language and Newness together are 40% of the score, per the weight column above, and all four are editing work, so they are where a page that has never been touched should start. Alignment is worth more than all of them combined, but it cannot be done from intuition: it needs the queries your buyers actually type, which is why it is a research task with an editing task attached rather than an editing task on its own.

Substantiation is last on this list for effort reasons and near the top on evidence. ConvertMate's GEO Benchmark Study 2026 reports that brands are 6.5 times more likely to be cited through third-party sources than through their own domain. That is a large multiplier on a dimension nobody can edit their way into, and it is the honest reason the framework caps it at 15% rather than weighting it by effect alone: a score is only useful if the thing it measures can be acted on.

What else do people ask about the SIGNALS framework?

What else do people ask about search signals and other frameworks?

What else do people ask about choosing a search signal framework?