# Cited Sources



A scan records **which domains the models cited** when they answered.

## What the List Tells You [#what-the-list-tells-you]

The list is a set of domains. Sorting them into groups is our habit, not a field
in the product, but what you can do about a domain depends on which group it is
in:

| Kind                           | Example                                 | What to do                                                                                  |
| ------------------------------ | --------------------------------------- | ------------------------------------------------------------------------------------------- |
| **Your own domain**            | `withsurface.com/platform/...`          | Good. Note which pages, and write more like them.                                           |
| **Review and directory sites** | G2, Capterra, category roundups         | Get listed, get reviews, get the description right. Usually the part you can edit yourself. |
| **Editorial and community**    | Blogs, newsletters, Reddit, Hacker News | Earned. Slow, and durable.                                                                  |
| **Competitor domains**         | A competitor's own comparison page      | They wrote the comparison and the model used it. Write yours.                               |

## The High-coverage, Low-visibility Case [#the-high-coverage-low-visibility-case]

You wrote the page and the model still does not cite you. The cited sources
usually explain it: the answer came from a roundup, a review site or a
competitor's comparison page, none of which mention you. The page is fine. The
model was reading other domains.

<Steps>
  <Step title="List the sources for the profile">
    `list_cited_sources` takes a `profileId` and returns the domains the models
    cited.
  </Step>

  <Step title="Get the evidence per prompt and model">
    `get_tracked_prompt_history` filters by prompt, category, model and date, so
    you can see which answer pulled from where, and when.
  </Step>

  <Step title="Separate the fixable from the earned">
    You can correct a directory listing this week. You cannot manufacture a
    community thread, and should not try.
  </Step>

  <Step title="Fix the fixable, then re-scan">
    Directory descriptions, category placement and review counts are the parts
    you control directly, so start there.
  </Step>
</Steps>

## Reading It as a Content Brief [#reading-it-as-a-content-brief]

The cited sources show what the model treated as a good answer. Two things
transfer into a brief:

* **The claims they make.** If three of four sources frame the category around
  per-seat pricing, address that head-on.
* **The shape.** Comparison tables, "best for X" statements and plain claims
  near the top tend to survive being quoted. Long preambles do not.

Hand the gap and the sources to the
[content agent](/platform/campaign-agents/content-agent) together.

## What This Is Not [#what-this-is-not]

Cited sources are not a backlink report. They show what a model reached for on
one question on one day. Domains move in and out, and absence from one prompt
says nothing about another.

[Acting on gaps →](/platform/ai-visibility/acting-on-gaps)