ALLMO tracks your visibility across the major AI assistants people actually use for information search. This article explains which models are supported, how they behave differently, and what those differences mean when you interpret your visibility data.
Supported AI models
ALLMO currently supports the following models for prompt monitoring:
ChatGPT (OpenAI)
Perplexity
Claude (Anthropic)
Gemini (Google)
Grok (xAI)
Mistral
You can enable any combination of these per report. Each enabled model runs your full prompt dataset on the configured schedule and feeds into the same Mention, Citation, and Share of Voice metrics.
Optionally, you can turn web search off, if you only want to analyze your visibility based on the model inherent training data.
Why model breakdown matters
Your visibility on one model does not predict your visibility on another. A brand can rank #5 on Perplexity, #9 on Gemini, and only #11 on ChatGPT for the same prompt set. The underlying reasons differ:
Each model has a different training data mix (what the model "knows" from pre-training).
Each model uses different retrieval and grounding sources at answer time (real-time web search, curated indexes, or no retrieval at all).
Each model applies different citation rules (Perplexity cites almost always, ChatGPT depends on mode).
Market share and user base differ by region and use case.
Reviewing visibility per model tells you whether a visibility gap is a brand awareness issue (missing in the training data) or a content authority issue (missing from retrieval sources).
How the major models differ
ChatGPT (OpenAI)
The most widely used AI assistant globally. ChatGPT has two relevant modes for visibility:
Without web search: answers come from the model's training data. Visibility here depends on whether your brand was prominent enough in the training corpus to be surfaced. Training data is cut off at a specific date, so recent brands are underrepresented.
With web search: the model runs a live query (see Query Fan-Out below) and cites real sources. The OAI-SearchBot crawler grounds answers in real-time web content.
GPTBot crawls content that may be used for future training. OAI-SearchBot handles real-time grounding for ChatGPT web search. Throttling either bot in robots.txt is a common cause of missing citations.
Perplexity
The most citation-forward model. Every answer includes a numbered source list, making citation tracking straightforward. Perplexity leans heavily on real-time web retrieval, which means well-structured, recent, and crawlable content performs disproportionately well. If a brand is strong on Perplexity but weak on ChatGPT, the likely cause is a training data gap rather than a content problem.
Claude (Anthropic)
Claude can be used with or without web search. ClaudeBot crawls public content. Claude tends to produce longer, more nuanced answers and is often used in professional and technical contexts.
Gemini (Google)
Gemini is tightly integrated with Google Search and often pulls from similar sources as traditional Google results. It has broad reach via Google AI Overviews, which share some ranking signals with classic SEO. Strong traditional SEO often translates into Gemini visibility, though not always.
Grok (xAI)
Grok is integrated with X (Twitter) and pulls heavily from that platform. Being active on X, or at least reposting LinkedIn content there, disproportionately affects Grok visibility.
Mistral
European-built model with a different training data mix than US-centric models. Can be relevant for European audiences and multilingual prompt sets.
Query Fan-Out: how search-enabled models work
For models with web search enabled, a single user prompt usually triggers multiple background search queries. This is called Query Fan-Out.
Example: a user asks "What are Germany's most important universities for entrepreneurial talent?" The model might issue three or more separate searches in parallel, for instance:
"German universities producing most startups spin-offs list 2024 2025"
"Stifterverband Gründungsradar 2025 top entrepreneurial universities Germany"
"WHU entrepreneurship startups WHU alumni startups"
The model then synthesizes one answer from all of those search results. This means:
A single prompt can surface citations from many different queries
Your content does not have to match the original prompt literally, it has to match one of the fan-out queries
Long-tail and entity-specific content often gets pulled in even when the user's prompt was broad
Query Fan-Out is a core reason why AI search is different from traditional SEO. Optimizing for one target query is no longer enough. Content needs to cover the adjacent questions a model is likely to generate on its own.
How to use model breakdown in ALLMO
Every ALLMO dashboard can be filtered by model. Typical workflows:
Prioritize: focus optimization on models with the largest user base in your market first (usually ChatGPT and Gemini, with Perplexity growing in B2B contexts)
Diagnose gaps: compare rankings across models on the same prompt set.
Track changes: AI models update frequently. Monitor per-model trends to catch sudden drops or gains after a model release.
Which models should you track?
At minimum, track ChatGPT (the market leader) and Perplexity (the most citation-transparent). Add Gemini if you have strong Google Search presence to defend, Claude if your audience is professional or technical, and Grok if you are active on X or target a US tech audience. Mistral is most relevant for European, multilingual, or privacy-sensitive use cases.
You can enable and disable models per report at any time.