Glossary: AI visibility & GEO

AI Mention Rate is the raw hit rate: across all queries, all providers and all daily runs in the last 30 days, how often did the brand appear in the AI answer?

It is the more accessible sibling of AI Prominence — without weighting for position, provider breadth, sentiment or recommendation strength. A brand can have a high mention rate (cited often) but a moderate AI Prominence (always at the end of the answer, neutral tone, never the top pick). The two metrics complement each other.

Google's AI Overview (previously "Search Generative Experience", SGE) is the AI answer Google has been rolling out above the classic search results since 2024–2025. Unlike with ChatGPT or Perplexity, the user sees both: the AI answer and the search-results list. In practice the AI answer often resolves the question and the click into the results list never happens.

For brands this has two consequences. First, previously reliable click metrics from Google Search Console become unreliable (see "AI Overview Signal"). Second, the choice of cited third-party sources directly determines which brands appear in the AI Overview — often those are different sources than the ones ranking in the classic top 10.

When a query racks up impressions in GSC but barely any clicks, that is a textbook symptom of Google's AI Overview: the answer sits above the organic results, the user gets it without clicking through.

Achtung.app scans GSC data for this pattern and surfaces the affected queries as AI Overview Signals. They are the high-priority targets for GEO work — places where organic visibility looks strong on paper but in practice gets absorbed by the AI card.

The signal is only visible when Google Search Console is connected to the brand.

AI Prominence is Achtung.app's headline metric. Six components feed in: Citation Frequency (20%), Citation Breadth across providers (20%), Recommendation Strength (25%), Position Score (15%) reflecting how early the brand appears in the answer, Sentiment of mentions (10%), and Trend Momentum over recent weeks (10%).

A score of 80 or above means Top-Promi (top-tier): the brand is cited reliably, on multiple AI providers, prominently and positively. Low scores are not a negative judgment by the AI — they simply mean the brand is not actively recommended by the model, which is exactly the gap GEO is meant to close.

AI Prominence is derived from brand-free queries (see "Organic AI Visibility"), not from explicit brand mentions.

In search-grounded answers, cited sources are a central lever. Whatever shows up there tends to show up in the AI answer, often as a linked citation. Achtung.app collects every third-party site AI models pull in for a brand's queries and ranks them by frequency, provider breadth, and opportunity.

The opportunity badge (High / Medium / Low) flags the sources that promise the biggest leverage: cited often, but where the brand doesn't yet have a presence. Press outreach, guest posts, or directory listings on high-opportunity sources are among the most direct ways to lift AI visibility.

Each tracked keyword is queried multiple times per AI provider per day. AI models are stochastic — the same question does not always produce the same answer. The Consistency Score measures how often the brand turns up across those repetitions: 100% means cited every run, 50% means cited half the time.

High consistency means the brand is firmly anchored in the model. Low consistency means the model knows the brand but does not reliably surface it — the typical state of brands on the edge of the recommendation set. The score is exposed per provider because consistency between ChatGPT, Gemini, and Perplexity often diverges sharply.

GEO stands for Generative Engine Optimization. Where classic SEO optimizes for ranking high in a list of links, GEO optimizes for being named, cited or recommended inside the answer itself.

That shifts the levers from keyword rankings towards three things: being reachable for AI live-retrieval crawlers, being recognizable in the models' training data, and having content the model can extract answers from. Classic SEO signals (content, authority, technical hygiene) still matter, but their weight and exact mechanics differ from pure SERP optimization.

Achtung.app measures GEO without prescribing how to do it: the tool tells brands where they stand in AI answers and which third-party sources the AI currently prefers. What brands do with that — content, PR, technical fixes — stays their decision.

A GEO gap is the most concrete call-to-action in the system. Instead of just saying "your score is 42", it says: "On this query the AI named competitors A, B and C, but not you, because…".

Gap reasons are standalone LLM analyses, not simple pattern matching. Typical reasons: no study or documentation on the topic exists on the site; the niche can't be unambiguously assigned; a local focus that the sources don't reflect; competitors with concrete numbers or comparison tables that the AI cites instead.

GEO gaps feed the weekly content suggestions: each suggestion names the cited third-party sources, the gap reason, and the structural element to publish (FAQ block, comparison table, dated benchmark).

The Joint Prominence Delta answers the question "does this brand belong more to classic web search or to AI answers?" on a normalized scale. Both prominence values are scaled to 0–1, then the formula subtracts the AI value from the web-search value.

A strongly negative delta means: the AI models recommend the brand disproportionately, even though it barely shows up in Google's result list (typical for rising SaaS players with good PR motion). A strongly positive delta means: the brand dominates classic web search but barely appears in AI answers (typical for entrenched incumbents with thin GEO work).

In industry reports the delta appears as the "Δ Joint Prom." column and as the sort key in the same-brand-asymmetry chart.

The AI world has two crawler families with very different functions. Live-retrieval crawlers grab pages now, because a user has just asked a question whose answer needs current information. Blocking them (in robots.txt or via an edge rule at the firewall) directly removes the brand from ChatGPT, Perplexity and Claude answers.

Current live-retrieval crawlers include OAI-SearchBot, ChatGPT-User, PerplexityBot, Perplexity-User, Claude-User and Claude-SearchBot. Achtung.app's weekly website-hygiene audit checks that all of them are allowed and that no silent edge block intervenes.

Training crawlers (see next entry) are something else, and the system deliberately does not grade them.

llms.txt is a proposed-but-not-yet-standardized file: a compact Markdown document at the host root that tells an AI model in one place what the site is about, what topics it covers, and which sub-paths hold the canonical sources.

Google's official guide for AI Overviews and AI Mode states that llms.txt has no effect on AI visibility. Achtung.app's website-hygiene audit still records its presence as an informational data point but, as of May 2026, no longer counts it as an "important" check.

Our own server logs back this up empirically: in the 30 days leading up to the publication of the companion data article on 17 May 2026, /llms.txt on achtung.app was fetched 90 times. The hits come from a single scraper using a spoofed Chrome user-agent (76 of 90), classic SEO and discovery crawlers (Barkrowler, AhrefsBot, BuiltWith, Dataprovider), and the odd profiling tool. Verified AI bots like ClaudeBot, GPTBot, PerplexityBot, OAI-SearchBot, Claude-SearchBot, ChatGPT-User, Perplexity-User, Applebot, Googlebot and Bingbot did not appear in this snapshot. The picture has since shifted slightly: in the 30 days to 1 June 2026, verified AI bots did fetch the file a handful of times, GPTBot twice (17 and 30 May) and OAI-SearchBot once (21 May). Every fetch came as part of a normal sitemap crawl, not a targeted pull, and stayed in the low single digits. A fetch is not a visibility effect, though: it does not change Google's assessment that llms.txt has no effect on AI answers.

Achtung.app itself ships an extensive llms.txt as a declarative entry point for agents that explicitly fetch that path, not from any expected crawl effect.

More in the methodology.

Classic brand analysis tends to ask AI models directly: "What do you know about Brand X?" That method measures nothing useful, because the AI will talk about any brand you name regardless of how visible the brand actually is.

Achtung.app inverts the question. The system runs daily brand-free queries from the brand's market ("What are the best karate schools in Berlin?", "Which tools help with project management?"). The query never contains the brand name or the domain. The AI either names the brand on its own — or it doesn't.

The result is a harder but more honest metric: organic AI visibility mirrors what a real customer experiences when asking the AI for advice.

The bare 0-to-100 number needs an anchor. The Prominence Tier provides it: four tiers borrowed from German celebrity language ("Promi" = celebrity) that hold on to the idea that AI visibility is about recognition inside a market.

The tiers are a UI aid; the actual value remains the 0–100 number.

Search-grounded means: the AI model does not answer from its training snapshot alone. It pulls live web content and cites it auditably in the answer. ChatGPT (with web browsing enabled), Gemini, Perplexity and Claude all offer this mode; Achtung.app currently tracks the three search-grounded providers ChatGPT, Gemini and Perplexity.

The mode is the foundation of serious AI-visibility measurement for three reasons: first, it mirrors what real users see; second, the cited sources are auditable; third, pure training-data answers systematically favor older brands that appeared in more crawls (closed-book bias).

In English product copy the term usually appears with the inline gloss live web search per query.

Training crawlers feed the pretraining of future model generations. Allowing them gives your content a chance to land in the next pretraining cycle for Anthropic, OpenAI, and Google models. Blocking them forgoes that lane but does not affect current visibility in live answers.

Allowing or blocking is therefore an editorial decision, not a hygiene flaw. Achtung.app reports the verdicts for GPTBot, Google-Extended, ClaudeBot, CCBot and Applebot-Extended in a dedicated row group inside the hygiene matrix, but it does not grade them and they never move the "AI-ready" badge.

Website hygiene is the technical baseline a brand needs to be reachable, parseable, and citable by AI assistants. Every Monday, Achtung.app fetches each tracked brand's homepage and records what it finds:

• AI crawler access: are OAI-SearchBot, ChatGPT-User, PerplexityBot, Perplexity-User, Claude-User, Claude-SearchBot allowed in robots.txt and not silently blocked at the edge (Cloudflare, WAF)?
• JSON-LD presence: which schema.org types appear on the homepage (Organization, WebSite, FAQPage, etc.)
• Reachability: HTTPS, sitemap, canonical URL
• Renderability: does the page produce readable content without JavaScript? AI crawlers do not execute JS.
• Metadata: Open Graph title and image used in AI answer previews

The output is a hygiene snapshot per brand and ISO week. Snapshots are append-only, so once enough weeks accumulate, longitudinal before/after analyses become possible.

Hygiene is a baseline measure of capture-readiness, not a ranking. It tells you what technical preconditions a brand meets; it does not directly predict citation count.