An AuditSpark.io whitepaper on why being found stopped being one job and became seven.
Website intelligence that sparks action.
TL;DR
- Visibility used to be one job: rank in a list of links. It is now seven jobs, and they fail independently. A site can pass any six and still be absent from the answer a buyer actually reads.
- The seven disciplines are Foundation, SEO, GEO, AEO, LLMO, SXO, and Authority & Trust. Each answers a different question, and each has its own failure mode.
- They form a sequence, not a menu: crawlable, then found, then cited, then chosen, then known, then converting, then trusted. Work done out of order is work done on a theory.
- Six of the seven can be read off your own website. One cannot. LLMO — whether engines actually mention and recommend you — lives inside the engines, and the only way to know is to ask them and record what comes back.
- The most common failure we measure is not a broken site. It is a site that is technically fine and structurally unquotable. On 40 real small-business sites, 67.5% had no question-shaped headings at all and 92.5% carried no answer schema.
- Readiness and visibility are different dials. Our own site scores 100/100 on AI/GEO readiness and 17.3/100 on measured AI visibility. That gap is not a contradiction. It is the whole point of separating them.
Start here: run a free AuditSpark.io audit to see your site scored across the ecosystem, and which discipline is actually costing you.
Executive summary
For twenty years, website visibility meant one thing: rank on the first page of Google. The audit industry grew up around that single goal. Keywords, backlinks, page speed, and meta tags all served one question — where do we appear in the list of links.
That question has not disappeared. It has been joined by six others, and the six are not variations on the first. When someone asks ChatGPT, Claude, Gemini, Microsoft Copilot, Perplexity, or Google's AI Overviews about a category, they do not receive ten links to evaluate. They receive a synthesized answer that names a few brands, cites a few sources, and silently omits everything else. Getting into that answer depends on things a ranking report was never built to check: whether the engine's crawler is allowed to fetch you, whether your page contains text shaped like an answer, whether anyone off your own site says you exist, and whether the visitor who does arrive can work out what you sell.
Each of those is a separate discipline with a separate failure mode, and they fail independently. This is the part most teams get wrong. They treat AI visibility as one more box on the SEO checklist, discover their rankings are healthy, and conclude they are fine. Then they run a probe and find that engines recommend three competitors by name and have never heard of them.
This paper lays out the seven disciplines, explains what each one measures and how it breaks, and — more usefully — separates them by how you can know. Five are scored by reading your own pages. One is classified rather than scored, because the underlying data does not support a number. And one cannot be read off your site at any resolution, because it does not live there.
The practical takeaway is that a traditional website audit is now roughly two-sevenths of the picture, and the missing five-sevenths is where most of the loss is. This is not a claim that SEO is dead. Google's own documentation keeps the fundamentals firmly in place. It is a claim that the fundamentals are now load-bearing for two audiences at once, and that several new gates have opened between a healthy site and a named recommendation.
Why it matters now
Three things happened at roughly the same time, and together they move this from a future concern to a present one.
Answer engines now carry real audience. OpenAI reported that ChatGPT reached roughly 900 million weekly active users in February 2026, up from about 800 million in October 2025, and the ChatGPT app crossed one billion monthly active users in mid 2026 by press accounts. Google has rolled AI Overviews and AI Mode into core Search, and the Gemini app is reported in the range of 750 million monthly active users. These are vendor and press reported figures and should be treated as directional, but the direction is not ambiguous. A large and growing share of buying research now passes through an answer layer that summarizes rather than lists.
The answer layer shows fewer brands than a results page. A traditional results page can surface ten organic links plus ads, giving a buyer many doors to walk through. A generated answer typically names a handful of options and cites a handful of sources. The visible surface area shrank, so the cost of being left out rose. Being on page one of links is no longer the same as being in the answer.
The failure modes multiplied faster than the tooling. This is the part that gets underplayed. It is not simply that there is a new place to be visible. It is that the ways to be invisible now outnumber the ways anyone is measuring. A blanket crawler block added years ago to stop scrapers. A page whose every heading is a marketing slogan rather than a question. A brand that exists nowhere on the open web except its own domain, so a web-grounded engine searching for third-party corroboration finds nothing. Each of those is invisible to a rank tracker, invisible to analytics, and fatal to a recommendation.
The urgency here is real, but it is not a reason to panic or to chase hacks. It is a reason to audit properly — meaning across all seven disciplines rather than the two most tools cover. The businesses that look closely now will find a small number of fixable problems before those problems cost them named recommendations in front of buyers.
Technical validation
This section separates what is established from what is still emerging, because credibility depends on that distinction.
Established: Generative Engine Optimization is a defined visibility problem
The academic origin of the term is the paper GEO: Generative Engine Optimization (arXiv:2311.09735), which defines generative engines as systems that gather and summarize information from multiple sources to answer a query, and which observes that content creators have little control over when and how their content appears in those answers. The authors propose methods to improve content visibility inside generated responses and report gains of up to forty percent in their benchmark setting.
The honest way to use this number is as evidence that visibility inside generated answers is a measurable, movable quantity, not as a promise that any tactic delivers forty percent for any site. The benchmark studied specific engines and content under specific conditions. The durable lesson is conceptual: inclusion in an answer is something you can influence, and therefore something worth auditing.
Established: SEO fundamentals still govern eligibility
Google's Search Central documentation on AI features states there are no additional technical requirements to appear in AI Overviews or AI Mode beyond being indexed and eligible to appear in Search with a snippet, and that you do not need to create new machine readable files or special schema to be eligible. Crawlability, indexability, useful content, page experience, and structured data that matches visible text remain the foundation.
Read carefully, this is the strongest available argument for treating the disciplines as a sequence rather than a menu. Google is saying that the Foundation and SEO disciplines are prerequisites for the AI surfaces, not alternatives to them. A site that is slow, unrenderable, or unindexed does not get to skip ahead to the interesting work.
Established: visibility now depends on which bots you allow
Major AI providers run more than one crawler, and they do different jobs. OpenAI documents GPTBot for training, OAI-SearchBot for surfacing sites in ChatGPT search features, and ChatGPT-User for user-initiated retrieval, and exposes independent robots.txt controls for each. Anthropic documents ClaudeBot for training, Claude-User for user-directed retrieval, and Claude-SearchBot for search quality, each controllable separately.
The implication is that "blocked" and "allowed" are not the two available states. A site can be fully allowed for training and fully blocked for the search crawler that actually feeds live answers, which is close to the worst of both worlds: the content contributes to a model and the brand gets no citation. This is the subject of Paper 2 and is introduced here to establish that access is now a deliberate configuration rather than a default.
Established, from our own measurement: most sites carry schema, but not the kinds engines lift from
We sampled 47 distinct hosts from real scan history, deduplicated by domain, and scored the 40 that were reachable. The sample is genuinely small-business-shaped — regional marketing agencies, engineering firms, law practices, wellness clinics — with a handful of larger outliers that, if anything, flatter the numbers.
26 of the 40 sites (65%) carried JSON-LD structured data that contained no answer-oriented type. They had Organization, WebSite, BreadcrumbList, LocalBusiness — the schema a website builder emits by default — and nothing an answer engine reaches for. A conventional audit checking "does this site have structured data?" returns a clean pass on all 26. The check is true and the conclusion is wrong.
⚠ Methodology caveat: n=40, homepage only, roughly ±7 percentage points at these rates. Treat these as indicative of the small-business segment we scan, not as a general population figure.
Emerging: content structure may influence citation behavior
A growing body of research argues that how content is structured — answer-first formatting, clear definitions, self-contained chunks — can influence whether and how generative engines cite it. One 2026 study held wording constant and varied only structure, reporting a lift in citation rates across six engines. This is promising and directionally useful, but it remains an active research area rather than settled fact, and we present it as a hypothesis to test rather than a law to obey. The safe, evidence-aligned move is to make content clearer and more quotable because that helps human readers and search engines too, not because any single formatting trick guarantees a citation.
Emerging: AI readiness files such as llms.txt
The llms.txt proposal describes a plain text file at a site's root that gives AI systems a curated summary of a site's key content. It is a community convention maintained through llmstxt.org, not a formal standard, and reported adoption sits in the low single to double digit percentages, concentrated among technical and documentation-heavy sites. Google has explicitly said no such file is required to appear in its AI features. We treat llms.txt as an emerging readiness signal worth understanding, not as a requirement and not as a guaranteed citation factor.
The seven disciplines
Here is the ecosystem. The order is deliberate and it is a dependency chain, not a ranking of importance.
| # | Discipline | The question it answers | How it breaks |
|---|---|---|---|
| 1 | Foundation | Can the page be fetched, rendered, and used at all? | Slow loads, layout shift, a mobile experience that fails, accessibility barriers that also block machine parsing |
| 2 | SEO | Can a search index find, understand, and rank it? | Thin or duplicated content, weak titles and headings, crawl and index problems, no topical depth |
| 3 | GEO | Is the site eligible to be cited by an answer engine? | Blocked AI search crawlers, missing or generic structured data, weak semantic HTML, no E-E-A-T signals, no third-party corroboration |
| 4 | AEO | Does the page contain text an engine can lift as an answer? | No question-shaped headings, answers buried in long prose, FAQ content with no markup, answers hidden in collapsed panels |
| 5 | LLMO | Do engines actually mention, cite, and recommend you? | Competitors named instead; brand absent; brand mentioned but misdescribed; visible only when the prompt already names you |
| 6 | SXO | Does the visitor who arrives understand and act? | Unclear value proposition, competing calls to action, weak messaging, a first impression that does not land |
| 7 | Authority & Trust | Is there credible evidence you are who you claim to be? | No real bios or credentials, no case studies, thin off-site presence, claims that cannot be checked |
A brief tour, because the acronyms do real work and are worth holding separately.
Foundation is the technical core: performance, mobile experience, accessibility. It is unglamorous and it gates everything. An engine that times out fetching your page has formed no opinion about your content.
SEO is the discipline everyone already has. It has not been replaced and its scope has not shrunk. What changed is that it is now a prerequisite for surfaces it was not designed to serve.
GEO — Generative Engine Optimization — is on-site citability. It asks whether your site is eligible to be quoted: are the AI search crawlers allowed in, is there structured data an engine can parse, is the HTML semantic, are there experience and expertise signals, and does any third party corroborate that you exist. It is scored 0–100 and it is entirely deterministic. No AI is involved in producing it, which is why we can give it away free on every audit.
AEO — Answer Engine Optimization — is the narrower and more mechanical question of whether the page actually contains liftable answers. Not "is this good content" but "if an engine wanted to quote two sentences that answer a buyer's question, are there two such sentences, are they attached to a heading that states the question, and are they marked up so the engine knows what it found." Paper 4 is devoted to it.
LLMO — Large Language Model Optimization — is the off-site half, and it is different in kind from everything above it. It cannot be read off your pages, because it is not a property of your pages. It is a property of the engines' behavior.
SXO — Search Experience Optimization — is what happens after the click: conversion, messaging, UX, brand clarity, visual design, first impression. Skipping it is how teams win demand they cannot convert.
Authority & Trust is E-E-A-T made concrete: credentials, evidence, and a presence beyond your own domain. It does double duty, since the same signals that persuade a buyer are the ones an engine uses to decide you are a real entity worth naming.
Why the order matters
The sequence is crawlable → found → cited → chosen → known → converting → trusted, and it describes dependency rather than priority.
This has a blunt practical consequence: diagnose from the bottom, but sell from the top. A client whose CDN is challenging every AI search crawler will produce a terrible LLMO report, and every conclusion you draw from it will be about their content when the actual cause is a firewall rule. Fixing content first would have been months of work aimed at the wrong discipline. Conversely, the finding that creates urgency in a sales conversation is almost always the LLMO one, because "three competitors are named and you are not" is concrete in a way "your semantic HTML is weak" never will be. Lead with the symptom; treat the cause.
Three kinds of measurement
This is the structural point of the paper, and it is the one that most changes how you should read any AI-visibility tool, ours included.
The seven disciplines are not equally knowable. They divide into three groups by how you can find out, and confusing the groups is the source of most bad analysis in this space.
Five are scored by reading your own site. Foundation, SEO, GEO, SXO, and Authority & Trust all yield a number because the evidence is right there in the page. You fetch the site, you check it against criteria, you get a score. It is repeatable and cheap, and two people running it independently should agree.
One is classified, not scored — and that is a finding, not a limitation. AEO gets reported as pass, partial, or gap rather than a number out of ten, and the reason is the data. On those 40 real sites, 67.5% had zero question-shaped headings and 92.5% had no answer schema at all. A weighted AEO score would read approximately zero for two-thirds of the population — a binary with a decimal point after it. It could not rank clients against each other, could not trend over time, and would tell most of the book the identical thing. Where a score cannot discriminate, publishing one is false precision, and the honest move is to say pass, partial, or gap and then name the specific missing thing.
There is a related trap worth naming, because it cost us a version to find. An early version of our own check counted any heading beginning with an interrogative word as a question. On real small-business copy, 44% of those matches were false positives — "Who We Are", "What We Do", "How We Help", "What clients say." These are declarative marketing headers that happen to open with a question word, and counting them inflates the score exactly where the site is weakest. A question requires a question mark. The lesson generalizes: a signal that looks reasonable in a spec can be systematically wrong on the population you actually serve, and the only way to find out is to measure against real sites.
One cannot be read off your site at all. LLMO is not a property of your website. It is a property of what six different engines say when a buyer asks a category question. There is no file to check, no tag to add, no score to compute from your own HTML. The only way to know is to ask the engines real questions and record what comes back — which brand names appear, which sources get cited, whether you are recommended or merely mentioned, and whether what is said about you is even true.
This is why readiness and visibility have to stay separate numbers. They measure different objects.
What that separation looks like when it bites
Our own site is the cleanest example we have, so we will use it rather than a flattering hypothetical.
AuditSpark scores 100/100 on AI/GEO readiness. Every crawler we want is allowed, the structured data is correct, the semantic HTML is clean, the E-E-A-T signals are in place. On the on-site half we have done the work completely.
AuditSpark scores 17.3/100 on measured AI visibility, with a share of voice of 0.0%. Across six engines, our brand appears in about one question in five, and when we strip out the mentions that do not count as independent recognition — questions that named us in the prompt, and answers where the engine cited our own domain back to us — nothing survives. The category is owned by Semrush, Screaming Frog, and Ahrefs.
A perfect on-site score and a near-absent off-site reading is not a contradiction, and it is not a broken instrument. It is the two dials doing their job. Our site is flawlessly structured to be cited and is barely being cited, because citation depends on material that exists off our domain — third-party coverage, community discussion, independent corroboration — and no further on-site work produces any of it. That is a content and distribution problem wearing a technical costume, and the only reason we can see it clearly is that the two numbers were never blended into one.
If a tool gives you a single "AI visibility score" that goes up when you add schema markup, it is measuring readiness and calling it visibility. Those are different objects, and the difference is exactly where the expensive mistakes live.
Business impact
The ecosystem framing is not an abstraction. It changes what shows up in the pipeline.
Lost demand you cannot see. When an answer engine omits you from a recommendation, there is no impression, no click, and no line in your analytics that says you were considered and passed over. Unlike a ranking drop, which shows up in your rank tracker, an exclusion from AI answers leaves almost no trace in the tools most teams already use. It is easy to assume everything is fine because traffic looks stable, while a growing slice of high-intent research quietly routes to competitors who got named.
Misrepresentation, not just absence. A subtler risk is being described inaccurately. An engine may state your pricing, your service area, or your specialty using stale or wrong information pulled from somewhere on the web. A confident, wrong answer about your business can do more damage than silence, and you will not know it is happening unless you look.
Effort aimed at the wrong discipline. This is the cost the ecosystem view is specifically designed to prevent. A team that reads "AI visibility" as a content problem writes content. A team that reads it as a technical problem adds schema. Both can be entirely correct in their execution and entirely wrong about which of the seven was actually failing — and neither will find out for a quarter, because the feedback loop runs through engines nobody is probing.
Why a ranking report is no longer enough. A standard SEO audit answers whether you can rank. It was never designed to answer whether an engine can reach you across the right bots, whether your page contains liftable answers, whether anyone off-site corroborates you, or whether engines name you when buyers ask. Those are four different questions with four different checks. A business auditing only the first is measuring a fraction of its visibility and calling it the whole.
The constructive framing is that almost all of this is fixable, and most of it is fixable on a schedule. The cost of inaction is not a catastrophe next quarter. It is a slow leak of named recommendations that compounds while you are not looking.
The agency and freelancer opportunity
For agencies, freelancers, and consultants, the ecosystem is not only a client risk. It is a new service conversation and a new line of revenue.
The old conversation was "we can improve your rankings" — familiar, commoditized, and increasingly hard to differentiate. The new conversation is "there are seven things that decide whether you get recommended, you are currently measuring two of them, and here is what the other five say about your site." That is a fresh reason to reach out to existing clients and a credible hook for new ones, because almost no small or mid-sized business has audited it.
The ecosystem also solves a real packaging problem. "AI visibility" as a single undifferentiated service is hard to scope and harder to price, because nobody — including you — knows what is in it. Seven named disciplines with distinct failure modes turn one vague engagement into a diagnostic and a sequence of scoped projects: an access and Foundation cleanup, a GEO and AEO structural pass, an Authority and off-site programme, a conversion review. Each has a clear definition of done, which is what makes it sellable at a fixed price.
Paper 6 is a full playbook for packaging, pricing, and selling this without overpromising. The point to register now is that early movers get to define this conversation with their clients before a competitor does.
Common mistakes
Assuming good rankings mean good AI visibility. They are correlated but not the same. SEO and LLMO are two disciplines with two failure modes. Treat them as two measurements.
Treating the seven as a menu. They are a dependency chain. Content work on a site whose search crawlers are blocked is work performed on a theory.
Blending readiness and visibility into one score. They measure different objects — one your site, one the engines' behavior. A single blended number hides exactly the case you most need to see, which is strong readiness with absent visibility.
Blocking AI crawlers by reflex. A blanket disallow added to stop scrapers also removes you from the search crawlers that surface you in answers. Decide deliberately which bots to allow for discovery and which to block for training.
Believing the hype that SEO is dead. It is not, and saying so undermines your credibility. Google's own guidance keeps the fundamentals in place. AI search raises the cost of weak fundamentals; it does not replace them.
Treating llms.txt as a magic switch. It is an emerging convention, not a requirement, and it does not guarantee citations.
Optimizing only for machines. A site perfectly readable to engines and confusing to humans wins demand it cannot convert. SXO is not optional.
Measuring once and moving on. Engine behavior shifts over time and varies across engines. A single snapshot of LLMO ages quickly. If it matters, measure it on a cadence.
The 30, 60, 90 day action plan
Days 1 to 30 — Foundation, SEO, and GEO access. Establish a baseline across all seven disciplines so you know which one is actually failing. Then fix access first: check robots.txt and any CDN or firewall rules to confirm you are not blocking the AI search crawlers, which are separate from the training crawlers and are the ones that feed live answers. Confirm your core content is in server-rendered text rather than locked behind client-side JavaScript. Access problems make every other improvement theoretical, so nothing else starts until this is clean.
Days 31 to 60 — AEO and Authority. Take your ten most commercially important pages and give each one a heading that states the question a buyer would actually ask, in question form, followed immediately by a self-contained answer of roughly 15 to 80 words. Mark up genuine FAQ content with the schema types answer engines look for, rather than leaving structured data at the builder default. In parallel, start the Authority work, which is slower and therefore needs to start earlier than feels necessary: real bios with checkable credentials, case studies with specifics, and a presence somewhere other than your own domain.
Days 61 to 90 — LLMO baseline and SXO. Run your first real visibility measurement: write down the ten questions buyers actually ask before choosing a vendor in your category, put them to several engines, and record whether you are mentioned, cited, recommended, or misdescribed — and which competitors appear instead. That is your baseline, and it is the only one of the seven you could not have obtained by reading your own site. Then review conversion, because everything above this line delivers visitors and SXO decides whether they matter.
Throughout, resist the temptation to work on all seven at once. The sequence exists because the dependencies are real.
Checklist
Use this as a fast self-assessment. Each unchecked box is a candidate for your action plan.
Foundation
- Core Web Vitals measured on real pages, not assumed
- Mobile experience genuinely usable, not merely responsive
- No accessibility barriers that also block machine parsing
- Core content present in server-rendered HTML
SEO
- Titles and headings clearly state each page's topic
- Content has real depth on the topics you want to be known for
- Indexing and crawl status verified, not assumed
- Sitemap present and current
GEO
- robots.txt reviewed for AI search crawlers, separately from training crawlers
- No blanket bot block at the CDN or firewall
- Structured data present and of types relevant to your content
- Semantic HTML, not a page of styled
divelements - At least one credible third-party reference to your organization
AEO
- Key pages carry headings phrased as real questions, with question marks
- Each question is followed immediately by a self-contained answer
- FAQ content is marked up, not just formatted
- Answers are not hidden in collapsed panels without markup
LLMO
- You have asked real category questions of at least two engines
- You know whether you are mentioned, cited, or recommended
- You know which competitors appear instead
- You have separated "found when named" from genuine independent recognition
- You have a plan to re-check on a cadence
SXO
- Above-the-fold message is clear within seconds
- A single, obvious primary call to action
- Forms are short and usable
Authority & Trust
- Real author or expert bios with checkable credentials
- Case studies, testimonials, or reviews with specifics
- A presence beyond your own domain
FAQ
What is the difference between GEO, AEO, and LLMO? GEO asks whether your site is eligible to be cited — crawler access, structured data, semantic HTML, credibility signals. AEO asks whether your pages actually contain liftable answers, meaning question-shaped headings with self-contained answers attached. LLMO asks whether engines actually mention and recommend you when buyers ask. The first two are read off your own site; the third can only be measured by asking the engines.
Do I need all seven disciplines, or can I pick the ones that matter to my business? They are a dependency chain rather than a menu, so the order matters more than picking favorites. Work on later disciplines is unreliable while earlier ones are broken — content restructuring on a site whose AI search crawlers are blocked cannot be evaluated, because nothing was ever going to read it.
Can a site score well on AI readiness and still be invisible in AI answers? Yes, and it is common. Readiness measures your own site; visibility measures engine behavior, which depends heavily on material that exists off your domain. Our own site scores 100/100 on readiness and 17.3/100 on measured visibility. On-site work cannot close an off-site gap.
Is SEO still worth doing? Yes. Google's documentation states there are no additional technical requirements to appear in AI Overviews or AI Mode beyond being indexed and eligible for a snippet, which makes SEO a prerequisite for the AI surfaces rather than an alternative to them. The accurate statement is that AI search raised the cost of weak fundamentals, not that it replaced them.
How often should AI visibility be measured? On a cadence rather than once. Engine behavior shifts over time and engines disagree with each other, so a single snapshot ages quickly and a single-engine check tells you about that engine rather than about your visibility.
How AuditSpark compares
If you're choosing a tool, these honest, side-by-side comparisons may help:
- AuditSpark vs Otterly — AI-visibility monitoring vs. an audit that tells you what to fix
- AuditSpark vs Semrush AI Visibility — a monitoring add-on vs. AI readiness and a full audit, included
- AuditSpark vs Profound — enterprise GEO vs. accessible website intelligence
- AuditSpark vs My Web Audit — a focused, automated audit vs. a full agency platform
- AuditSpark vs Website Audit AI — a 15-category audit vs. a free visual snapshot
See where your own site stands
Run a free AuditSpark AI & GEO Readiness audit — score, executive summary, and the fixes that matter, in minutes.
Run a free audit →Next in the series → AI Crawler Access