Everything you need to understand AI search visibility — the research behind it, how to fix it, and what actually moves the needle.
SIGNALS is the AI-visibility (AEO) framework, the intelligence platform that measures it, and the practice that does the work.
If your pages aren't showing up when buyers ask ChatGPT or Perplexity for recommendations, these resources explain why — and exactly what's involved. When you're ready to fix it, request a free visibility assessment.
What an engagement covers: per-engine measurement across ChatGPT, Google AI Overviews, Perplexity and Claude, the page work that makes claims extractable, and the re-measurement that shows what moved.
See what's involved → FrameworkAll 7 dimensions, their weights, and the research behind each one. Why alignment dominates, what grounding means, and how to score higher on each signal.
Explore the framework → ResearchThe peer-reviewed studies behind every SIGNALS weight. 2M+ citations analyzed across 4 papers — how we know what actually drives AI citation frequency.
Read the research → GuideWhy your website is invisible to ChatGPT, Claude, and Perplexity — and what AEO is. How AI citation works, how it differs from SEO, the four-stage retrieval pipeline, and the 7 signals that determine whether AI cites you.
Read guide → GuideThe 6 most common reasons AI engines skip your pages — and exactly how to fix each one. Includes a diagnostic checklist and before/after examples.
Read guide → ChecklistA complete checklist of the 25 structural signals that determine whether ChatGPT, Claude, and Perplexity cite your page. Based on 2M+ citation analysis.
View checklist → How-toChatGPT reads from the Bing index, not Google's — so a page that ranks on Google can be invisible in ChatGPT. The five steps to get retrieved, indexed, extracted, and cited.
Read guide → How-toAI Overviews name a few sources before the buyer scrolls — and it's a different game from ranking #1. How to become one of them, grounded in a captured real Overview.
Read guide → How-toPerplexity runs its own crawler and index, cites the most sources of any engine, and prizes freshness. How to be one of the sources it names — including the query fan-out most guides miss.
Read guide → How-toClaude's web search runs on Brave, not Google — and it cites fewer, more carefully chosen sources than any engine. Harder to earn, worth more. How to become one.
Read guide → How-toGemini grounds its answers in a live Google Search and attaches the chunks it used to the sentences it wrote. What that means for your pages, and why Google-Extended is a licensing decision rather than an SEO one.
Read guide → How-toYour AI visibility is four numbers — one per engine — and they rarely agree. How to measure share of voice the way it actually behaves, query by query, across repeated runs.
Read guide → DefinitionThe plain definition: how often AI engines name your company when buyers ask — four numbers, one per engine, that decide the shortlist before a buyer makes contact.
Read definition → DefinitionThe GEO term, its 2024 academic origin, how it differs from AEO and SEO, and what the research shows actually moves citation in generative answers.
Read definition → ExplainerReddit is one of the most-cited domains in AI answers, and the published figures for how much disagree wildly. What the studies actually measured, and what a business should copy rather than chase.
Read explainer → Off-site signalsAlmost every tool ChatGPT names has a review profile, and the engines almost never cite one. Why three 2026 studies appear to conflict, and what that leaves worth doing.
Read explainer → Off-site signalsDomain authority accounts for roughly 3% of the variation in AI citation probability, and brand mentions out-correlate links by about three to one. What the published numbers report, and where a link budget converts into citations instead.
Read explainer → Off-site signalsAbout 85% of brand mentions reaching AI systems originate on pages you do not own, which is a real finding and a bad reason to start there. Why owned pages come first, and how one measurement run tells you which side your problem is on.
Read explainer → ExplainerThe five factors behind every AI citation — retrieve, extract, trust, match, corroborate — and what each one means for whether your page gets named.
Read explainer → ExplainerOne prompt becomes roughly nine to eleven hidden searches before an answer is written, and most of them have no search volume at all. What the published measurements say, and how to make one page eligible for several sub-queries.
Read explainer → MeasurementBrand mentions in AI answers halve roughly every 31 days, and 73.5% of citations appear once and never return. The four metrics that measure decay, how often to sample them, and why a stable domain can hide total page churn.
Read guide → BuyingOne named owner on the marketing side, with contributors in content, web and PR who do not hold the number. What the owner is accountable for each month, and what fragments when nobody is.
Read guide → Business caseThe request gets funded when it reads as revenue already at risk rather than as a new channel to try. The four numbers a CFO will accept, where each comes from, and how to size the first ask.
Read guide → BuyingStructural work stays done and everything downstream of it does not, because the answer is regenerated on every ask. Which items are one-off, which recur, and how to run a pilot that produces a readable result.
Read guide → ExpectationsRetrieval responds in days and durable citation takes a quarter, which is why the timelines quoted online contradict each other. What has actually been measured, and what nobody has.
Read guide → Engine behaviourDays on a site the crawlers already visit often, weeks on one they visit rarely, and being quoted is a separate event later. Which OpenAI crawler decides whether you appear, and how to tell crawled, indexed and cited apart.
Read guide → Engine guideCopilot grounds its answers in the Bing index, and it is the one engine that reports your citations back to you. How to read the AI Performance report, and what to fix first.
Read guide → Technical AEOAhrefs found 97% of llms.txt files got zero requests in a month, and Google says no AI system reads the file. What the four published studies measured, and the one case where publishing one still makes sense.
Read guide → Failure modeOnly 27% of the phone numbers AI assistants gave out matched the business's own profile. Two different failures produce the same symptom, and one question asked twice tells you which one you have.
Read guide → TeardownA real-world teardown of a page that executes AI citation so well that Claude consistently cites it as a neutral source. Signal by signal breakdown with proof.
Read teardown → Case StudyWe ran SIGNALS on our own site and scored 28/100 — Critical. Real before/after screenshots, every fix documented, and exactly why each one worked — the same done-for-you work we do on a client's pages.
Read case study → ExplainerHow ChatGPT, Claude, and Perplexity actually retrieve and cite content — the four stages, where most pages fail, and what to fix first.
Read explainer → GuideNot named when buyers ask ChatGPT, Perplexity, Claude, or Google's AI? It's almost always one of five specific, fixable reasons — and rarely a quality problem. How to tell which is yours.
Read guide → How-toAI can't cite a page it can't read. A practical four-check test for whether ChatGPT, Perplexity, Claude, and Google's AI can reach, render, and extract your pages — and what to fix if they can't.
Read guide → How-toChatGPT names a few firms when someone asks for a lawyer — and it reads from Bing, not Google. The four steps to make your firm one of the names, for the high-value case types worth taking.
Read guide → ComparisonAEO is new enough that anyone can claim it. The criteria that separate a real AI-visibility practice from a repackaged SEO shop — with a buyer's criteria table and the questions to ask before you sign.
Read guide → ComparisonBuy a tracking tool, build it in-house, or hire a done-for-you agency? A side-by-side comparison of what each actually covers, what it costs in time, and when each one makes sense.
Read guide → ComparisonProfound, Otterly AI, and the DIY trackers vs. a done-for-you agency — real pricing, honest pros and cons, and who each is actually for. Where SIGNALS fits, and which category you need before you spend a dollar.
Read comparison → ComparisonWordPress, Webflow, Squarespace, Wix, Shopify and custom builds scored on the four things that decide whether an engine can read and cite your pages, plus a ten minute test for the site you already have.
Read comparison → Off-siteWikipedia is one of the most cited domains in AI answers, and that is not the same as being the company an engine names to a buyer. What an entry actually does, who qualifies, and the corroboration that works instead.
Read guide → Failure modeA crawler cannot fill in a form, log in or buy a subscription, so it reads the landing page and nothing behind it. What the paywall citation data shows, and how to keep the leads without losing the citations.
Read guide → MeasurementGA4's AI Assistant channel, the utm_source=chatgpt.com tag and a free-text field on your form each catch something the other two miss. The three layer setup, and the four numbers worth reporting.
Read guide → MigrationsEngines cite a specific URL and a specific sentence, and none of them offer a coverage report. What to baseline before launch, which pages to protect, and how long recovery actually takes.
Read guide → BuyingForrester put 94% of B2B buyers using generative AI in their most recent purchase, 55% of them to compare vendors. What that justifies spending on, and how to check your own category instead of guessing.
Read guide → MeasurementNo published benchmark transfers, because the denominator is whatever prompt set the publisher chose. The two comparisons that do hold, and how much movement is just noise.
Read guide → Technical AEOSpeed behaves like a gate rather than a dial: it decides whether the page is read, not whether it is used. What to measure, what to fix, and what to leave on the human-performance backlog.
Read guide → BuyingA measurement run, a page-fix queue and a publishing backlog, wrapped around one baseline. What month one looks like, what your team has to supply, and what is usually not in scope.
Read guide → GuideSoftware categories are unusually well served by their own vendors, which is why a mid-sized SaaS company can win citations a larger domain loses. Which pages earn them, and the order to fix things in.
Read guide → BuyingWhat a handover can destroy is the measurement history, not the citations. The four artifacts to get in writing before you give notice, how long to overlap the two suppliers, and why the number moves the week you switch.
Read guide → MeasurementFour numbers on one page, each with its method printed beside it. Why a single percentage is not reportable, how to handle published figures that disagree with each other, and what to leave off a board slide.
Read guide → BuyingThree tests decide it, and the first one is whether the measurement is real. What happens to your citations if you stop, which parts of the work you can take in-house, and how to run the renewal conversation.
Read guide → GuideMove the buyer out of New York and most of the shortlist changes. What language and location each decide, which pages to localise first, and how to measure a market you do not sell from yet.
Read guide → GuideNo controlled study shows that a hidden price costs you citations, and most of your competitors publish nothing a machine can read. What an engine does with the gap, and what counts as a readable price.
Read guide → GuideAn engine picks passages, not one URL per site, so two pages that each half-answer a question can both be read and neither quoted. How to spot the split in your own run logs, which page to keep, and how to merge it without losing the citation.
Read guide → BuyingThirty days proves whether the supplier measures honestly. It does not prove a citation outcome, because cited pages average nearly three years old and the measurement is noisy at short horizons. Four pilot shapes, and what each one can settle.
Read guide → BuyingA few days of engineering at the start, a couple of hours a month after. Which fixes genuinely need code, which are CMS edits a marketer can make, and why client-side rendering is the one blocker nothing else gets past.
Read guide → GuideLinkedIn is the second most cited domain in Semrush's dataset, and the citation credits a linkedin.com URL. What gets quoted, what that is actually worth to you, and why the answers still belong on your own domain.
Read guide → GuideNo page count earns citations, and word count correlates with being cited at 0.04. One page per buyer question, each answering it outright. Which five to write first, and how to tell whether content is the bottleneck at all.
Read guide → BuyingFour things are cheap for you and expensive to buy: crawler access, text that renders without JavaScript, your real buyer question list, and one named approver. Why schema markup is not on the list, and what to deliberately leave alone.
Read guide → BuyingNo, and Google's own documentation says indexing and serving are not guaranteed. What an agency can legitimately commit to, what accepting a guarantee costs you, and the five things to ask for instead.
Read guide → BuyingLow keyword volume is usually a measurement artefact. A missing brand footprint is the constraint that bites. How to test whether the assistants recommend anyone in your category before you budget for it.
Read guide → MeasurementThe formula alone can double a number: the same 5,018 answers return 26.72% visibility and 11.87% mention share. The four causes behind the gap, and the five questions to ask before you compare two dashboards.
Read guide → MeasurementOften not much of it. Being selected as a source and shaping the written answer are separate events, a cited page carries roughly four times more of a ChatGPT answer than of a Perplexity one, and a citation count summed across engines adds unlike things.
Read guide → BuyingThe review queue is the constraint, not the rules. The measured page-level levers are wording and ordering changes that rarely create a new regulated claim, so split the backlog in two and ask legal to approve a pattern once.
Read guide → BuyingA partner taking the order does not take you out of the contest. Brand mentions run at 2.8% on GPT when your own domain is absent from the retrieval path and 49.0% when it is cited, and in our own sourcing study suppliers' own pages carried 73% to 89% of citations.
Read guide → BuyingLarge sites do not need every page fixed. 87% of the variation in citation counts sits between pages of one site, the site-wide schema rollout has no measured within-site effect, and the work is selection and wording on a few dozen pages.
Read guide → GuidePersonalisation is the smaller of the two reasons. Cited sources overlap at a Jaccard of 0.32 to 0.43 when the same prompt is re-run inside a day, so most of what looks like targeting is sampling. How to measure around both.
Read guide → GuideNobody has measured subdomain against subfolder for AI citations, and the claim that engines discount subdomains is asserted rather than tested. Use a subfolder for new content, leave an existing subdomain alone, and spend the migration budget on wording.
Read guide →The SIGNALS framework is grounded in peer-reviewed research. These are the four papers that inform every dimension, weight, and recommendation in the SIGNALS assessment.
2M+ citations analyzed across 4 engines. Found prompt-content alignment (β=+0.37) is the dominant predictor — above domain authority and all structural signals.
Read paper ↗10,000 queries tested. Adding statistics increased AI visibility by 41%, source citations by 40%, expert quotes by 28%. The foundational peer-reviewed study.
Read paper ↗First taxonomy of AI citation failure modes: retrieval, parsing, ranking, generation. The study reports 40%+ citation improvement by targeting the actual failure stage.
Read paper ↗ConvertMate's benchmark study covered 12,500+ queries across 8,000 domains. 83% of AI citations come from outside Google's top 10. Heading hierarchy present in 68.7% of cited pages.
Read paper ↗Start with a free visibility assessment — it shows exactly which of the 7 citation signals your pages are failing across the four engines. From there, SIGNALS does the work: the BLUF rewrite, buyer-language FAQ, H2 restructuring, and JSON-LD schema, implemented on your pages. The research section explains the peer-reviewed evidence behind each fix type.
SEO optimizes for Google's PageRank algorithm: backlinks, domain authority, click-through rate. AEO optimizes for AI retrieval pipelines: vocabulary alignment, structural clarity, sourced statistics, quotable content. According to ConvertMate's GEO Benchmark (2026), 83% of AI citations now come from pages outside Google's top 10 — meaning strong SEO rankings no longer predict AI citation. You need both, and they require different fixes.
Alignment — making sure your page vocabulary matches how buyers actually search — carries 35% of the SIGNALS score and is the only signal with a documented causal effect independent of domain authority (Discovered Labs, 2026, β=+0.37). Adding sourced statistics is the second biggest lever: Princeton GEO (KDD 2024) found it increases AI visibility by 41%. Both are central to the work we do for you — Alignment is where we start.
AI index refresh typically takes 2–4 weeks after a page is updated. Structural changes (BLUF, FAQ, schema) are picked up faster than vocabulary changes, which require re-indexing. Once we ship the fixes, your SIGNALS score reflects the improvement immediately, and AI citation frequency changes within 2–6 weeks as crawlers re-process the updated content.
A free, PULSE-powered visibility assessment maps exactly where you’re cited and where you’re invisible — against the competitors winning your category, category by category.
Request a free visibility assessment →