Skip to main content

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.

Labels you attach to prompts. Once attached, they flow through the entire dashboard as filters, so you can slice visibility, citations, and reports by any dimension that matters to your business. Think of them as the seasoning that turns one big visibility number into a useful set of smaller ones.

​
Walkthrough

​
Examples

Attributes are whatever you want them to be. Some common shapes:
  • Stage of buyer journey: awareness, consideration, decision.
  • Audience type: SMB, mid-market, enterprise.
  • Use case: ergonomics, productivity, aesthetics.
  • Geography or market when one project covers multiple regions.
  • Anything specific to your client’s business that you want to be able to slice on.
If you have a dimension you’d report on in a slide deck, it should probably be an attribute.

​
Built-in vs custom

OpenLens ships a small set of built-in attributes that get assigned automatically. Everything else you define yourself.
  • Built-in:
    • Branded: assigned to prompts that name a brand (your own or a competitor) in the question itself. See the worked example below.
    • Language: one tag per language present in your project (English, Italian, Spanish, etc.), assigned by the language the prompt is written in.
  • Custom: anything you add at the project level. Engine, audience, journey stage, geography, whatever you want. No real limit on count.
Built-in attributes cannot be archived. Custom ones can.

​
How attributes get assigned

Four paths a prompt can pick up attributes:
  • During onboarding. You list the attributes you care about up front. OpenLens auto-generates prompts that try to fit them, and labels each generated prompt with the attributes it fits best.
  • Inline when editing a prompt. Open any prompt in the Prompts tab, type a new attribute in the attribute field, click the plus (creates the project-level attribute if it doesn’t exist), then Save. Useful when you discover a useful slice mid-flight.
  • In Settings. Project-level attribute management for bulk add, edit, and archive.
  • Via the MCP server or REST API. Same actions exposed programmatically. See the MCP server and REST API.
Manual tagging in the dashboard is tedious. That’s what MCP is for. Drive the labeling from Claude (or your favorite agent) once and you’re done.

​
How attributes flow through OpenLens

Once a prompt has an attribute, that attribute is available as a filter everywhere downstream:
  • Dashboard: filter the visibility score by attribute. Compare your awareness-stage SOV against your decision-stage SOV.
  • Citations: see which sources get cited for which slice of your prompt library.
  • Reports: include per-attribute breakdowns in PDF exports.
  • MCP and REST API: every read endpoint that returns scores or mentions accepts an attribute filter parameter.
The practical implication: every attribute you add is one more axis you can use to explain a result to a client.

​
Filter modes

The dashboard’s attribute filter has three modes you can mix:
  • All of: a prompt has to have every selected attribute to pass. Use this to narrow down to an intersection (branded AND Italian).
  • Any of: a prompt only needs one of the selected attributes. Use this to union (Italian OR English).
  • Hide: exclude every prompt with the selected attribute. Use this for negation (hide branded to see only unbranded).
Mix the three to land on the exact slice. A filter preview shows the prompt count under your current selection before you commit, so you can sanity-check the math. Worked example. Overall visibility on a sample project was 68.1%. Filter to branded-only (prompts that name the brand) and visibility bumped to 72.3%. Italian-only was higher still. Italian AND English gave zero prompts because no prompt is in both languages. Italian OR English gave the sum. Hide-branded dropped visibility to 60%. Same dataset, four different views, all real.

​
Example: branded vs unbranded

The clearest example of why attributes matter. A prompt can name a brand in the question itself or not. OpenLens ships this as an attribute on every prompt.
  • Branded prompts name your brand or a competitor in the question. Example: ‘Is Acme better than Roost for desk setups?’
  • Unbranded prompts describe a need without naming any brand. Example: ‘Best laptop stand for a standing desk.’
Visibility on branded prompts will always run higher than on unbranded ones. You’re asking the model about yourself, so it answers about you. That’s the question being measured. Higher numbers there mean exactly what they say.
  • Branded SOV answers: when someone asks about us by name, what does the model say?
  • Unbranded SOV answers: when someone asks for the category we live in, do we even come up?
Both matter. They answer different questions about the same brand. Branded is the floor: if it flatlines to zero, models don’t know your brand by name. Unbranded is harder to move and more directly tied to whether anyone discovers you in the first place. Filter the dashboard by the branded/unbranded attribute to see each in isolation.

​
Lifecycle

  • Add an attribute in project settings, or directly when editing a prompt.
  • Edit an attribute’s value on a prompt by opening the prompt and changing the attribute.
  • Archive a custom attribute. Marks it inactive for new prompts. Existing prompts keep the historical label so reports stay readable.
  • Unarchive. Same place, reverses the archive.
Built-in attributes (language) cannot be archived.

​
How to use it

  • Pick two or three attributes that match the way you talk about your client’s market. Add them up front.
  • Audit your prompts after the first run. Anything mis-labeled? Fix it.
  • Apply attribute filters on the dashboard to validate that your slices look like the business.
  • When a client asks ‘how are we doing for [audience X]?,’ filter by that attribute and have the answer in one click.
  • If the manual tagging is taking too long, drive it from Claude via MCP. The MCP tools cover add, remove, archive on attributes and edit on individual prompts.