Evidence over vibes

Publishing more pages is not the same as getting cited more. Original information, real experience and accurate facts correlate with AI visibility; page count barely does.

Four things changed at the engines this month — a publisher-payment pilot, a Google content-guidance update, a new UGC data pipeline, and GPT-6's interactive answers. Here's what each actually means.

A sourced roundup of the AI search statistics worth repeating in 2026 — each with its publisher, sample and date, and what the number does not mean.

A results page ranks pages and lets you choose. An AI answer retrieves, selects and writes a few sentences that name a few brands for you.

Most wasted AI-visibility spend traces to a short list of errors: one run treated as proof, chasing rank instead of frequency, blocking the wrong bot, and a few more.

Our opinion: most teams need a baseline and a fix list before they need a dashboard. Here's the honest decision, including the case against us.

Put the answer in the first paragraph, ask the question in the heading, and stop writing once one question is answered. That's most of it.

Bing was the first major engine to give publishers free citation data — total citations, cited pages, and the grounding queries behind them. Here's what it shows and what it doesn't.

Bing now shows you your own AI citations for free. Google shows impressions, not clicks. Here's what each first-party report covers, and where third-party tools fill the gap.

Publish your canonical facts once, in the same words, in the same places — and repeating that consistency is most of what you can actually control about AI accuracy.

AI answers about you start from what's findable under your own name. A clean homepage, claimed profiles and healthy branded search are the base layer everything else sits on.

AI engines don't make up your pricing — they repeat it from somewhere. Usually a comparison blog, a review listing, or your own old page. Here's how to find and fix each one.

List every source an AI engine cites for your competitors but never for you. That list is your content and outreach plan — the GEO equivalent of a keyword gap analysis.

Citation share is the slice of an answer's sources that are actually yours — not how often you're mentioned, and not the same thing as share of voice.

Mentions across credible sources correlate with AI visibility more strongly than backlinks do. Earn them with real data and expert quotes. Never buy them.

Say who you are in the same words everywhere, and make it checkable — on your own site and on the third-party sources an AI system also reads.

Several popular GEO tactics have thin or contradicted evidence. Here's what Google's own guide rules out, what a 75,000-brand study actually found, and what's still untested.

There's no single citation algorithm. Retrieval, relevance, freshness and source type each play a part — and some of them are yours to influence.

AI knows a brand from memory built during training, or from pages fetched live. You can influence the second one quickly and the first one only slowly.

Each engine documents a little about how it finds sources and leaves most of the rest unpublished. Here's exactly what's confirmed, by whom, and what still isn't.

Anthropic, which makes Claude, runs three separate bots for training, search indexing and live user fetches. Each is controlled independently in robots.txt.

Gemini sits beside Google's search products, not inside them. Here's what Google documents about how it grounds answers in live search, and what's simply not published.

Perplexity retrieves live and always shows its sources, which makes it the easiest engine to check — and the one whose results track Google's rankings most closely.

AI Overviews and AI Mode both draw on Google's core search systems, but they cite different sources for the same question far more often than they agree.

Buyers don't type fragments into AI chat. They ask full, specific, often messy questions — and the type of question matters more than how many people search it.

Perfect AI attribution doesn't exist. Teams that are measuring anything real rely on proxies: self-reported source, branded search lift, direct traffic, and sales-call mentions.

Pick your real rivals, ask them the same question you ask yourself, run it enough times, and write down exactly how you did it.

A good prompt set mirrors real buying questions at every funnel stage, branded and unbranded, worded several ways — not a dozen variations on your own brand name.

A score is a summary, not evidence. Before trusting any AI visibility report, check the prompts, run counts, dates, actual answers and sources behind the number.

Most GEO claims online aren't evidence — they're one person's one run. Here's the method: a testable hypothesis, matched pages, enough runs, and a disclosed limitation.

A clean robots.txt doesn't guarantee an AI bot can actually read your page. JavaScript-only rendering, a CDN's bot toggle and a firewall rule can each block access on their own.

Four studies rank content types differently because they define 'content type' differently. Here's what each measured, and what to publish despite the noise.

Five numbers, one chart, a finding, a fix list, a next step. That's the whole page — anything more and the boss stops reading.

AI engines can paraphrase what everyone already knows. They can't invent a number nobody has published yet — and that gap is where a small team's own research gets cited.

Update facts, dates, examples and sources on your best pages, and log every change. Freshness matters most for engines that retrieve live, like ChatGPT search and Perplexity.

Search bots and training bots are different jobs with different bot names. Allow what you want found, and decide training separately — one robots.txt rule doesn't control both.

One checklist across nine layers, strategy to measurement, with an honest evidence rating on every item — not a promise that the list is complete.

Keep your SEO foundations, add answer structure, then build third-party authority. The order matters more than which label you put on the work.

Schema markup makes a page's facts machine-readable. A controlled test found it doesn't move AI citations on its own, and Google says no special markup is needed for its AI features.

Before anything else, run pass-or-fail checks on crawler access and site structure. These 11 checks are the same ones behind ContextOwl's own report.

AI answers often lean on reviews, roundups, communities and listings as much as or more than a brand's own pages. Plan for the whole source ecosystem, not just your site.

GA4 can show some visits from AI engines through a native channel and a custom channel group. It misses copy-paste visits, app traffic and most AI Overview clicks — treat it as a floor, not a full count.

There's no 40% that's good everywhere. The only honest baseline is your own competitor set, run the same way, on the same day.

AI search engines split one question into several related searches before answering. Coverage of those hidden sub-questions decides who actually gets retrieved.

You can't watch every engine daily with a small team. Start where your buyers actually ask, write content that travels, then check each engine on its own terms.

Three studies get quoted as one 'AI is killing clicks' story. They measured three different things. Here's what each one actually shows.

Four services cover the work: an audit, content and technical fixes, third-party presence, and reporting clients can actually read.

Ten pass-or-fail checks, each with a clear bar to clear, that you can run by hand in an afternoon — no software required.

One-line definitions for the terms this blog uses most, each linked to where we cover it in depth.

Your share of voice is your slice of every brand mention across a fixed set of prompts and runs. Mix engines or prompt sets between measurements and the number stops meaning anything.

A bare score hides what it's built from. An honest score discloses its prompts, engines, run counts and date, and sits next to the evidence instead of replacing it.

A brand can be named without being cited, cited without being recommended, or recommended with no source at all. Measuring 'AI visibility' as one number hides which of these actually happened.

Not in any sense that survives a second run. Presence across repeated runs is the honest unit to chase; position is mostly noise dressed up as a number.

Before an AI engine answers a shopper's question about your brand, it has already pulled from your marketplace listing, your reviews and maybe a Reddit thread — not just your website.

A proposed markdown map of your site for AI agents. Evidence that major AI crawlers actually read it is thin, and Google says it doesn't use it at all.

Five labels, one underlying job: get named, described and cited correctly wherever someone is searching for an answer.

In May 2026, Google told SEOs what its AI features reward and what to stop doing — but the guidance covers only Google's own systems, not ChatGPT, Perplexity or Claude.

ChatGPT can answer from what it learned during training or from a live web search it runs on demand. Only the second path involves choosing sources at all — and that choice leans heavily on Bing.

Thirty minutes, five to eight prompts, three runs each, and a spreadsheet — that is the entire manual method, before any paid tool enters the picture.

Picking an AI visibility tool comes down to four things: which engines it checks, how many times it checks them, what evidence it shows, and how it prices the answer.

Four metrics survive repeated testing and one does not — here is how to track the four and quit reporting the one.

No. Clicks fall on queries where an AI summary appears, and that's real and measured — but SEO fundamentals are what AI features are built on, and the KPI stack is what actually needs to change.

Position inside an AI answer barely repeats; how often you show up at all is the number that holds steady — and the one worth reporting.

The same prompt, asked twice, almost never returns the same brand list. Appearance frequency across many runs holds up far better than where a brand lands.

GPTBot, OAI-SearchBot and ChatGPT-User are three separate bots with three separate jobs — block the wrong one and you can quietly remove yourself from AI answers.

AEO means writing so a direct-answer box or an AI engine can quote you. It is a real, older discipline with a new generation of surfaces — not a separate field from SEO.

AI search visibility is how often, how accurately and in what company AI engines name, describe and cite your brand. It measures the answer, not the ranking.

GEO is the practice of making a brand easier for AI engines to find, understand and cite. The research behind it is scoped, not a guarantee, and Google says that for its own features this is still SEO.