Citations

The Citations page shows how often AI answers name your brand and point to your website, how that compares with your competitors, and which sources AI cites in the answers where a competitor appears and you don't. It's where you find out why AI is recommending certain brands — and which sources are doing the work.

For background on what citations are and why they matter, read Citations and sources.

What you see

Everything is on one page. An old link or saved view that opened the former Sources tab now opens this page and scrolls down to the sources card.

Every answer in the window is put in one of four states for the brand being measured:

Named and referenced — the answer names the brand and references its website.
Named only — the answer names the brand but doesn't reference its website.
Domain only — the answer references the website without naming the brand.
Neither — no name and no website.

1. Five headline numbers

Domain referenced — the share of answers that reference the brand's website (named and referenced, plus domain only). Underneath, once the competitor ranking has loaded, it names the brand that leads that ranking and its rate, or the brand that comes next when the leader is you.
Named or referenced — the share of answers in any state except Neither. Underneath is the change, in percentage points, between the first half of the window and the second.
Named, not cited — the share of answers that name the brand without referencing its website, and how many answers that is.
Neither — the share and number of answers with neither.
Answers analysed — how many answers the numbers rest on, and how many days in the window had answers.

Under the tiles, one bar splits all of those answers into the four states.

The change splits the window's days at the middle and compares the named-or-referenced rate in the later half with the earlier half. Each half's rate is worked out from its own answer counts. When the window has an odd number of days, the later half gets the extra one. If either half has no answers, no change is shown.

2. Domain reference rate over time, and against competitors

On the left, Domain reference rate over time draws the rate for each day that had answers. Days without answers are hatched: they're gaps in the data, not zeros. Once the ranking loads, the best-placed competitor appears as a flat line at its rate over the whole window, labelled "(window)", because competitors have no day-by-day figures here. A competitor whose rate is zero isn't drawn.

On the right, Domain reference rate vs competitors lists every tracked competitor's rate, with the measured brand placed at its rank and highlighted. The top seven are shown, plus the measured brand if it ranks lower, and a line under the list sums up where it stands against the leader. The Competitors link opens the Competitors page with the same tags, and the same window when Competitors offers it.

3. Sources cited when you're missing

The sources AI cites in answers that name a tracked competitor but not your brand, starting with the source found in the most of those answers. Switch between Domains and Pages to see whole sites or individual URLs. Each row shows:

Gap answers — how many of those answers cite the source. An answer counts once per source, however often it cites it.
Prompts — how many prompts those answers came from.
Competitors cited with it — the competitors named in those answers, the models that cited the source and when it was last seen.
A tag: Third-party for sites where you can earn a mention, or Competitor site for a tracked competitor's own website.

The first six rows show at first; the Show the top … link under the list opens the full list, which holds up to the top twenty-five domains or fifty pages. The note at the bottom says how many of the competitor-only answers had sources that could be extracted.

Start action on a row opens a short form to turn that source into a tracked action (for example, "Get cited by g2.com"). You'll find it afterwards in the Actions view of the Recommendations page.

This card always looks for your own brand, even when you choose a competitor under Brand measured.

4. By prompt

Every prompt in the window, one row each: how many answers it had, a bar of the four states, Cited (named and referenced, plus domain only) and Neither. You can sort the list (least visible first is the default), search it, and page through it with Previous and Next.

Click a prompt to open its evidence: the answers behind it, which you can narrow to one state. Select an answer to read its full text and the sources extracted from it.

Filters and controls

The controls sit beside the page title:

Coverage note — when the data was captured and how many days in the window had answers. Open it for the exact window, the number of answers, what the change compares and how each number is worked out. Its dot turns amber when fewer than half the days in the window have answers.
Filters — Model, Intent, Category, Brand measured and Tags. They apply to every card. Choose a competitor under Brand measured to see the same numbers, trend and prompt list for that competitor.
Date window — from 7 to 365 days. The default is the last 30 days.
Views and Copy link — every filter is kept in the page's address, so a copied link reopens the same view. Save this view… in the Views menu keeps it in your saved views.
Export CSV — every prompt in the list for the current filters, not just the rows on screen. CSV export starts on the Starter plan. See Exports and API.

What to do with this data

Three concrete plays:

Play 1: Work through the sources cited when you're missing

Start at the top of Sources cited when you're missing and look at the Third-party rows. Each is a site AI already trusts for these questions, and it's citing your competitors instead of you:

Pitch the publication for coverage.
Ask to be added to their comparison post (most editors update on request if you're a credible fit).
Ask happy customers to leave reviews on platforms (G2, Trustpilot) where you're under-represented.

Use Start action on the ones you take on, so the work and what happens afterwards are tracked.

Play 2: Turn "Named, not cited" into citations

Named, not cited counts answers that mention you but don't point to your website. AI knows you, but it's leaning on other sources to describe you. If that share is high:

Your content may not be matching the language of customer prompts.
You may not have the kind of content AI prefers (comparison guides, use-case pages, customer stories).
Your site's authority signals (backlinks, recency) may be weak for your category.

Sort By prompt to find the prompts with the most named-only answers, and open one to read what AI said and which sources it used.

Play 3: Look for old pages in the Pages view

Switch the sources card to Pages. If AI keeps citing an old review or a years-old comparison post for your competitors, they may not be earning new citations either. A fresh, better piece from you has a real chance of replacing it.

What this page doesn't show

Direct attribution. AI doesn't always reveal every source it used. Citations represent what's surfaced in the response, not the model's full training-data reasoning.
Click-through. A citation is a signal, not a click. Whether a user clicks through depends on the answer wording.
Authority weighting. Sources are ranked by how many answers cite them, not by how authoritative they are. You'll need to apply judgment.
Like-for-like domain matching. Competitors' rates match their domain in the answer text; your own brand's rate is read from stored source links marked as your domain. The coverage note says so too.
Every source. Not every answer has sources that can be extracted. The note under the sources card says how many did.

Plan availability

The Citations page is on every plan, including the free trial. CSV export starts on the Starter plan.