AI Visibility · SEO, AEO and GEO
Most of what is sold as "AI visibility" has no evidence behind it. We will show you the studies — including the ones that say our own industry is wrong.
Two thirds of searches now end without a click. Your next patient or client is asking an assistant before they open a browser, and about three businesses get named in the answer. This is the work that puts you among them, and it is not the work most agencies are currently billing for.
Our AI Visibility service covers all of it — SEO, AEO (Answer Engine Optimization) and GEO (Generative Engine Optimization) — run as one program rather than sold as three.
SEO · AEO · GEO
SEO, AEO and GEO, defined plainly
Three labels, one job: making sure a machine can find you, understand you, and say your name when someone asks it for a recommendation. We run all three as a single AI marketing program.
SEO — Search Engine Optimization
Classic organic and local search. Still the largest source of trackable traffic for a local service business, and still the index most AI answers are built on top of.
Rankings, local pack, technical health, content.
AEO — Answer Engine Optimization
Being the source an answer engine quotes: AI Overviews, voice assistants, featured snippets. The question is no longer where you rank — it is whether your name is inside the answer.
Entity accuracy, reviews, directories, quotable copy.
GEO — Generative Engine Optimization
AI search optimization for generative assistants — ChatGPT, Perplexity, Gemini, Claude. Sometimes called LLM SEO or ChatGPT SEO. Earned third-party presence carries most of the weight here.
Earned coverage, server-rendered HTML, cross-platform measurement.
What actually changed
Put those together and the shape of the problem is clear. Your rankings can be identical to last year while your traffic falls by a third, and it is not a penalty and your SEO vendor did not break anything. The results page is answering the question itself, and a growing share of people are asking an assistant before a search engine ever gets involved.
The practical question is no longer where you rank. It is whether the machine says your name.
The uncomfortable part
This is a brand-new field, and most of it is being sold on faith
AEO, GEO and LLMO are used more or less interchangeably by practitioners, and there is no settled consensus on what any of them formally means. What there is, increasingly, is controlled research — and a lot of it contradicts what is currently being sold. Here is where the evidence sits today.
| Tactic | Evidence | What the research actually found |
|---|---|---|
| Earned third-party presence reviews, directories, forums, video, editorial coverage |
Strongest | Across an analysis of more than 23,000 citations, roughly three quarters of the citations that mention a brand come from somewhere other than that brand's own website. Wikipedia, Reddit and YouTube appear repeatedly among the most-cited domains on every platform studied. (Omniscient Digital; platform citation breakdowns from Digital Bloom, Profound and Ahrefs. Exact percentages vary meaningfully between studies — the direction does not.) |
| Server-rendered HTML content present without JavaScript |
Strong | AI crawlers are simple HTTP fetchers, not headless browsers. Analyses of crawler logs across GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and others report no JavaScript execution. If your content only exists after a script runs, those crawlers do not see it. (Multiple crawler-log analyses, 2026. Architecture-consistent and low-risk regardless.) |
| Brand mentions including unlinked |
Good | An analysis of 75,000 brands reports that brand mentions correlate with AI search visibility considerably more strongly than backlinks do. (Ahrefs, 2026, as reported. Google has separately stated that artificially manufactured mentions are filtered by its existing spam systems — so this is earned coverage, not a link-building scheme.) |
| Freshness and original data | Good | Content cited by AI systems is reported to be meaningfully more recently updated than content ranking organically, and answer-first structure with quotable original statistics and clear inline citations recurs across studies as a positive signal. (Ahrefs citation analysis; GEO benchmark research. Specific percentage lifts vary by study and should be treated as directional.) |
| Platform-specific work | Good | Only around one domain in ten is cited by both ChatGPT and Perplexity for the same query. Perplexity's answers overlap heavily with Google's top ten; ChatGPT's barely overlap at all. A single-platform strategy does not transfer. (Per-engine citation audits, 2026) |
| Schema markup sold as an AI citation lever |
Null result | 1,885 pages that added JSON-LD schema were tracked against roughly 4,000 control pages using difference-in-differences analysis. Adding schema produced no meaningful uplift in AI citations on any platform, and a small statistically significant decline of 4.6% in Google AI Overviews. Google's own May 2026 guidance states structured data is not required for AI search visibility. (Ahrefs, May 2026, corroborated by Search Engine Roundtable) |
| llms.txt | Not read | Google's John Mueller has stated publicly that no AI system currently uses llms.txt, and Gary Illyes confirmed Google does not support it. One AI-visibility vendor measured 62,100 AI bot requests to a site with a correct llms.txt file: 84 of them — one tenth of one percent — actually fetched it. A separate analysis of 300,000 domains found no correlation between having the file and citation volume. (Google statements; Otterly fetch-log data; SE Ranking correlation study) |
| Content chunking and "machine-friendly" rewrites | Unnecessary | Google's published guidance states there is no requirement to break content into small chunks for AI systems and no need to rewrite content to sound machine-friendly. (Google guidance, May 2026) |
| Prompt injection and cloaked AI-specific pages | Do not do this | No evidence of benefit, and serving different content to AI crawlers than to humans is cloaking — a spam violation with real penalty and reputational risk. (Tactic classification reviews, 2026) |
| Google Business Profile posts | Unstudied | We could find no controlled study testing whether GBP posts affect AI or local AI citations. We will not claim they work and we will not claim they are useless. They are cheap, so we do them as part of profile hygiene, and we do not bill them as an AI visibility tactic. |
Why we publish this at all
Because you are going to be pitched llms.txt files and schema packages by somebody this quarter, and the best available research says both are close to worthless for the thing they are being sold for. We would rather lose a deal to honesty than win one and spend six months billing you for a file that nothing reads.
It also tells you something about how we will behave when a campaign is not working.
What we actually do
Six workstreams. Every one of them maps to something in the evidence column above.
One — Machine access and rendering
We verify that AI crawlers can actually reach and read your pages: server-rendered HTML so content exists without JavaScript, clean robots directives that name the current AI user agents explicitly rather than relying on a wildcard, a working XML sitemap referenced from robots.txt, correct canonicals, and fast server response.
This is unglamorous and it is the single most common thing we find broken — including on the websites of firms selling AI services.
Two — Entity and data consistency
Assistants assemble an answer about you from fragments scattered across the web. Name, address, phone, hours, practice areas or procedures, provider and attorney credentials, and service descriptions have to agree everywhere they appear — your site, your Google Business Profile, the vertical directories, the data aggregators and the review platforms.
Contradictory data does not produce a wrong answer. It produces no answer, because the model reaches for a source it trusts more.
Three — The earned surface program
The highest-leverage work, because roughly three quarters of what a model says about you comes from off your own site. Review velocity and review text that names the procedure or practice area and the city. Presence in the specific directories your vertical's answers are built from. Honest participation in the forums and communities that get cited. Video. Local press and genuine "best of" placement.
Earned, never manufactured. Google filters inauthentic mentions, and we will not build you a liability.
Four — Answer-first content architecture
Pages built to be quoted: the answer stated directly near the top, original data and statistics a model can lift, clear inline attribution, and a question-and-answer structure that maps to how people actually ask. Built per service and per jurisdiction or per city, because that is the granularity the questions come in.
We still implement schema — for the rich results that remain and for entity clarity. We do not sell it as an AI lever, because the controlled study says it is not one.
Five — Conventional search, still
Local pack and organic still produce the majority of trackable traffic for local service businesses, and Google's position is that AEO and GEO are, for Google Search, still just SEO. Profile optimization, correct primary category, location and service pages, citation integrity, internal linking and technical health all continue.
Anyone telling you to abandon SEO for AEO is selling a rebrand, not a strategy.
Six — Measurement, honestly framed
A GA4 channel group that isolates referral traffic from ChatGPT, Perplexity, Gemini, Copilot and Claude. Sampled prompt tracking across the major assistants every month, run on the question set your buyers actually use. Citation-source mapping so you can see which third-party surfaces are feeding your answers. Server-log review of AI crawler activity.
With the caveat below, stated on every report.
The measurement caveat we put on every report
No tool on the market reads real user query logs from OpenAI, Google, Anthropic or Perplexity. Every AI visibility platform works by sending sample prompts and recording what comes back. Model outputs are non-deterministic and personalized, so any "share of voice" number is a sampled estimate that will vary between runs and between tools.
GA4 has its own limit: interactions inside AI mobile apps and in-app browsers frequently pass no referrer at all, so referral tracking systematically undercounts AI-driven traffic rather than overcounting it.
We report these as directional trends, tracked consistently over time. Anybody presenting you a precise AI market-share percentage is describing their sample, not the market.
What this looks like in your industry
Dental
A study of 824,997 citations from AI responses to health questions, collected across five platforms between April and July 2026, found that commercial and platform sources took 75.4% of citations. Academic journals took 8.3%, government and institutional sources 3.4%, and hospitals and providers just 1.0%.
That is the opposite of what most people assume about health queries. Assistants are not defaulting to .gov and .edu for everyday health questions — they are pulling from commercial health content, platforms, video and community discussion. (LLM Pulse, 2026. The study discloses that its dataset is weighted toward the topics its participating brands operate in, so treat the commercial share as an upper bound rather than a precise figure.)
For a practice, the practical implication is that your visibility is built on review text, healthcare directory presence, video, community discussion and local coverage — not on adding schema to your services page.
Legal
A 2026 legal AI visibility report found that seven attorney directories — Chambers, Legal 500, Super Lawyers, Best Lawyers, Martindale, Avvo and Justia — dominate citation share for legal queries, with individual firms rarely cited as independent voices. (5WPR / Haute Lawyer, 2026. The report's methodology was not published, so treat the finding as directional.)
If that holds even approximately, the strategy for a firm is not to out-publish the directories. It is to be correctly and completely present inside them, with the profile depth, practice-area specificity and review volume that makes you the firm the directory surfaces — and then to build the earned coverage that gives an assistant a second, independent reason to name you.
Every campaign, landing page and profile we build is reviewed against ABA Model Rules 7.1 through 7.3 and your state's equivalents before it goes live.
Questions we get
What is Generative Engine Optimization (GEO)?
GEO is the work of getting a business named and cited inside generative AI answers — ChatGPT, Perplexity, Gemini, Claude and AI Overviews. In practice it means earned third-party presence, accurate entity data, server-rendered content and consistent cross-platform measurement. Some vendors call the same work LLM SEO or ChatGPT SEO.
What is Answer Engine Optimization (AEO)?
AEO is optimizing to be the source an answer engine quotes rather than a link a user has to click. It grew out of voice search and featured snippets, and today it covers AI Overviews and assistant answers. We run SEO, AEO and GEO as one AI visibility program, not as three invoices.
Is AEO different from GEO? Do I need both?
Practitioners use AEO, GEO and LLMO more or less interchangeably, and the tactics overlap almost completely. GEO came out of a 2023 academic paper; AEO predates it and originally described voice search and featured snippets. Google's own position is that for Google Search these are still just SEO. We use "AI visibility" because it describes the outcome rather than staking a claim in a terminology argument that has no settled answer.
How long does this take to work?
Two to three quarters for the earned and content work to compound. Technical access and entity consistency can be fixed in weeks and sometimes produce movement quickly. Anybody promising you results in thirty days is either doing something risky or describing something they cannot measure.
Can you guarantee we show up in ChatGPT?
No, and neither can anybody else. Models are non-deterministic, results are personalized, and no vendor has access to the ranking mechanism. What we guarantee is the work, the measurement and the honesty about what the measurement can and cannot tell you.
Our SEO agency says schema is the key to AI visibility.
Ask them for the study. The best available controlled research — 1,885 pages tracked against 4,000 controls — found no meaningful citation uplift from adding schema, and Google's own guidance says structured data is not required for AI search visibility. We still implement schema, because it earns rich results and it clarifies entities. We just will not tell you it is the lever.
Should we block AI crawlers to protect our content?
That is a real decision with a real trade-off and it deserves a conversation rather than a default. Blocking training crawlers and blocking search crawlers are different choices with different consequences — and blocking Google-Extended does not remove you from AI Overviews, because those draw on the standard search index. For a practice or a firm whose problem is being found, blocking is almost always the wrong trade.
Do we have to stop doing normal SEO?
No. Local pack and organic still produce most of the trackable traffic for local service businesses, and the signals overlap heavily. Anyone telling you to abandon SEO for AEO is selling you a rebrand.
Find out what the machines currently say about you.
Twenty minutes, free. We will run your question set live on the call, show you what comes back, and tell you honestly whether the gap is worth closing. You keep the findings either way.
Sources cited on this page. SparkToro, In 2026, Less Than One Third of Google Searches Still Send a Click (June 2026), using Similarweb clickstream panel data. Pew Research Center, AI Overviews click-behavior study (July 2025), 68,879 searches from 900 U.S. adults. BrightLocal, Local Consumer Review Survey 2026. Ahrefs, We Tracked 1,885 Pages Adding Schema. AI Citations Barely Moved (May 2026), corroborated by Search Engine Roundtable. Ahrefs brand-mention analysis of 75,000 brands (2026). Omniscient Digital, citation source analysis (23,000+ citations). Google Search guidance on AI search tactics (May and June 2026). John Mueller and Gary Illyes public statements on llms.txt. Otterly.AI llms.txt fetch-log measurement. SE Ranking llms.txt correlation study (300,000 domains). Per-engine citation audits on ChatGPT and Perplexity overlap (2026). LLM Pulse, AI health citation analysis, 824,997 citations across five platforms (April–July 2026). 5WPR / Haute Lawyer, 2026 Legal AI Visibility Report. Google developer documentation on FAQ and HowTo rich result deprecation (2023–2026).
Where a study's methodology was not published, or where figures differ between sources, we say so at the point of use. Percentages in this field vary considerably between studies and should be read as directional. We re-verify every claim on this page against its source twice a year.