The Analyst

The CiteMetrix Analyst is an AI assistant that has read your data — your scans, your hallucinations, your competition tracking, your ModelScore breakdown — and can answer questions about it in plain English. Instead of clicking through dashboards trying to find a specific insight, you can just ask.

The Analyst lives in two places: a slide-out panel attached to most dashboard pages on the web, and a dedicated tab in the PWA where it’s the primary way you interact with your data on a phone or tablet.

What the Analyst can answer

The Analyst is grounded in your real data, not generic AI knowledge. Useful question patterns:

• “What changed in my ModelScore this week?” — pulls the trend, identifies which component moved, and explains the contributing scans or hallucinations.
• “Why did Perplexity stop citing us?” — looks at recent Perplexity scans, finds the inflection point, and surfaces hypotheses (often: a hallucination got introduced, a content change broke schema, or a competitor took over a query).
• “Which hallucinations should I fix first?” — sorts your active hallucinations by leverage (severity × frequency × cross-platform reach) and gives you a prioritized list with recommendations.
• “Compare us to [competitor name] on AI visibility” — side-by-side share-of-voice across platforms, plus the keywords and platforms where the competitor is winning.
• “Summarize this week’s changes” — a digest of the most material movements across all your tracking.

What the Analyst is not useful for:

• General AI knowledge questions (“what’s the capital of France”). The Analyst is scoped to your CiteMetrix data — for unrelated questions, use ChatGPT or Claude directly.
• Real-time external research. The Analyst doesn’t browse the web; it works from your scan history. If you need to know what’s happening right now on a platform, run a scan first and then ask.
• Predictions that aren’t grounded in trend data. The Analyst can extrapolate observed patterns; it can’t tell you the future of AI visibility for your category.

Where the Analyst appears

On the web dashboard: Look for the Analyst icon in the top-right of most pages — ModelScore, Citation Scans, Hallucinations, Competition. Click to open a slide-out panel. The Analyst is page-aware — opening it from the Hallucinations page primes it with the context of the hallucinations you’re looking at, so questions like “which of these should I fix first?” work without you having to specify which list.

In the PWA: The Analyst gets its own tab in the bottom navigation. On mobile it’s the primary surface — a chat-style interface optimized for voice input (tap the mic icon and ask). Voice input is supported on tablets and larger phones (viewport > 768px); below that it’s text-only.

Conversation persistence: Both surfaces save your conversation history. Pick up a thread on the web that you started on your phone, or vice versa. Threads are scoped to the domain you’re currently viewing, so questions about Domain A don’t bleed into Domain B’s context. (The web dashboard currently shows threads only inline — a unified history viewer is on the roadmap.)

How the Analyst gets its answers

When you ask a question, the Analyst:

1. Routes the question to a coordinator that decides what data to pull. (Your ModelScore? Recent hallucinations? Competition data? All of the above?)
2. Pulls the relevant data from your database
3. Sends the data plus your question to Anthropic’s Claude — the same family of models powering the rest of CiteMetrix’s AI work — with instructions to answer based on what’s there
4. Streams the answer back to your screen

A few consequences of this design:

• Answers are grounded in observable facts. When the Analyst says “your ModelScore dropped 8 points this week,” it’s looking at actual snapshot rows. You can ask “show me the data behind that” and it will surface the scans.
• The Analyst won’t hallucinate findings. It can summarize, prioritize, and reason about what’s there, but it can’t invent data that isn’t. If you ask about something the Analyst hasn’t been given access to, it’ll say so.
• Each conversation costs API calls. The Analyst uses your configured Anthropic key (or the agency fallback if you don’t have one). A typical conversation runs $0.05–$0.20 worth of API spend. Not free, but cheap relative to the time saved.

A typical workflow

The Analyst is most useful when you have a specific question rather than when you’re exploring. Compare:

Less useful: “Tell me about my data.” More useful: “What’s the biggest change in my ModelScore over the last 30 days, and what caused it?”

Less useful: “What should I do?” More useful: “I’ve got 30 minutes. Which two hallucinations would have the biggest impact if I fixed them today?”

The Analyst will respond to vague questions, but the answers tend to be vague back. Specificity in, specificity out.

Voice input on the PWA

If you’re using the PWA on a tablet or larger phone:

• Tap the microphone icon in the Analyst input field
• Speak your question
• The transcribed text appears in the input — review it, edit if needed, then tap send

The voice transcription never auto-submits. You always confirm before the question goes through. This is intentional — voice transcription has accuracy issues, and you don’t want a misheard query running up API costs or surfacing wrong data.

Voice input uses the browser’s native Web Speech API, which means it works offline-ish (transcription happens locally on most devices) and doesn’t require a separate service.

What the Analyst doesn’t do

• It doesn’t take action on your behalf. It can recommend; it doesn’t execute. Mark a hallucination resolved, run a scan, update a brand fact — those still happen via the regular UI. (Adding agentic actions is on the long-term roadmap, but the bar for “an AI does things in your account without explicit approval” is higher than the bar for “an AI answers questions about your data.”)
• It doesn’t replace dashboards. For at-a-glance status, charts and metrics are still faster than asking. The Analyst earns its place when you’re investigating something specific or when you want a summary that pulls from multiple panels at once.
• It’s not your support channel. Questions about how CiteMetrix works belong here in the docs; questions about your bill or subscription belong to eric@citemetrix.com. The Analyst will sometimes try to answer those questions and may get them wrong.

Next steps

• If you haven’t used the Analyst yet, open it from the dashboard and ask: “Summarize the last 30 days for this domain.” — it’s a low-stakes way to see what kind of summaries it produces.
• Install the PWA if you’d like to ask the Analyst from your phone.