Dashboard deep-dive

Documentation Index

Fetch the complete documentation index at: https://docs.openlens.com/llms.txt

Use this file to discover all available pages before exploring further.

Reading the dashboard well takes a few mental models. This page explains what the numbers actually mean, what can move them without anything real changing, and what to do about a bad result.

​

The four metrics

The dashboard shows four numbers. Get the definitions right and the dashboard reads itself. Visibility is the percentage of platform responses where your brand was mentioned at all. If 100 prompts ran across 7 platforms (700 responses) and your brand appeared in 280 of them, visibility is 40%. This is the mention rate. Mentions is the raw count behind visibility. 280 in the example above. Same data, different presentation. Share of voice is your mentions as a fraction of all brand mentions (you plus every competitor you’re tracking) for the same set of prompts. If you got 280 and your three competitors got 420 between them, your SOV is 40%. Visibility tells you how often you appear. Share of voice tells you how often you appear compared to who else is in the conversation. Average position is your rank in the answer when the AI lists multiple brands, averaged across every response that mentions you. Lower is better. 1.0 means you’re first every time. 1.4 means first most of the time and second the rest. Position answers ‘when the model mentions us, how prominently?’ rather than ‘how often does the model mention us?’ Visibility and position can disagree. You can be mentioned less often than a competitor but ranked higher when you are mentioned. Both numbers are real and they tell you different things.

​

Per-platform visibility

Same visibility number, broken out by AI platform. For each platform you also see your gap to the top competitor on that platform and a status indicator. Read it like a scorecard. ChatGPT can love you while Perplexity ignores you. The status indicator gives you a one-glance summary per platform, so you can spot the disagreements worth acting on.

​

Visibility trendline

Time series of your visibility number over a date range. Appears once you have at least two completed runs. Set up recurring runs in Settings (weekly or daily) and the line populates automatically. On-demand runs add a point every time you trigger one. Use this to spot real movement vs single-run noise. One run flipped against you isn’t a trend. Wait for three in a row before you act on it.

​

Competitor snapshot

The headline view of how you stack against the brands you’re tracking. Three numbers per brand: visibility %, share of voice, and average position. Sorted by share of voice. Position and share of voice can diverge. A competitor can win on share of voice (mentioned more often than you) while you win on position (ranked higher when you are mentioned). Both numbers are real. They answer different questions: who shows up most, and who looks best when they do. A brand sitting at position 1.4 means models name it first most of the time even when other brands are in the same answer.

​

Topic performance

Visibility broken out by topic. Click into any topic for the per-prompt breakdown underneath. Use this view to find the topics where competitors dominate. Those are the gaps.

​

AI sources

The list of cited domains models referenced when answering your prompts. The closest thing to a backlink graph for AI search. Each row is one cited domain, with:

• Count: how many times the domain was cited across the run.
• Platforms: which AI platforms cited it.
• URLs: the specific pages the models pointed at, with title and per-URL count.

Filter by topic, platform, or prompt attribute. Default view is the top 20 cited domains, up to 10 URLs each. Read this view to figure out where the models are pulling their answers from in your space. If review sites dominate citations for unbranded prompts in your category, that’s where you need to show up. The competitor blogs you keep seeing in answers about your problem space are content gaps you can close one by one. Roadmap: domain categorization (own brand, competitor, news, UGC, review site, third party) and category-based filtering. Not in the UI yet.

​

Slice by attribute

Every interesting dashboard cut is just an attribute filter. The most prominent example is branded vs unbranded prompts (prompts that name a brand vs prompts that describe a need without one). Filter the dashboard by attribute to see each slice in isolation. For the full mechanics, see Prompt attributes.

​

What moves the visibility score (for real reasons)

• Your brand is being mentioned in more responses than before.
• Competitor mentions dropped.
• A specific topic moved up or down because models are surfacing different sources for it.

​

What moves the visibility score (for non-reasons)

• You edited prompts mid-run. Different questions get different answers.
• The active platform mix changed.
• Run-to-run variance. AI responses are stochastic. The system is stable in aggregate but any single prompt’s answer can flip.

Knowing which is which is the difference between acting on signal and acting on noise.

​

Why platforms disagree

ChatGPT, Perplexity, Gemini, and the others don’t return the same answer for the same prompt. They’re trained differently. They search differently. Some lean heavily on cited sources, some lean on training data. Disagreement is the baseline.

​

Honest about noise

Run-to-run variance for any single prompt can be substantial. The system is stable in aggregate over a project’s full set of prompts. Don’t read a single prompt’s flip from one run to the next as a real change. Look at the trend over several runs before deciding something has actually moved.