Citations & Sources

Citations & Sources is where you see which URLs AI platforms actually pulled from when answering your prompts. Under the four summary cards, Your category’s sources at a glance sums up where your category’s answers get their sources and whether you’re gaining ground on your tracked rivals. Below it the page splits into two halves: Citations, the sources cited where your brand appeared, and How your category gets answered, the domains that anchor your category’s answers — competitive intelligence, clearly labelled by whether presence there can be pursued at all.

​

Citations

Every citation is a source URL an AI platform used or linked to in a response where your brand was mentioned. Like the Mentions feed, citations are grouped by prompt and organised into recency sections — Last 30 days, Previous 30 days, and Older. Four summary cards sit at the top of the page:

• Unique sites — how many distinct websites were cited in answers that mentioned you over the last 30 days. A site counts once however many of its pages, or however many scans, cited it; the caption underneath counts the distinct pages.
• Citation events — every appearance of a source in an answer, counted once per appearance per scan, over the last 30 days. It measures how often sources surface, not how many there are, so it grows with how often your prompts run.
• You vs rivals — your share of source co-occurrences in your latest scan: the co-occurrences involving you, out of all those involving you or one of your tracked rivals. At a glance explains what a co-occurrence is, and charts this share scan by scan.
• Unique Prompts — how many distinct prompts produced your citations.

Expand a prompt to see every domain cited for it, broken down by country. Expand a specific source to open the response drawer with the citation highlighted in the AI’s actual text, so you can see exactly where and how it was used. The Branded / Non-branded filter from Prompts is also available here, narrowing the citation list and summary cards to prompts matching that slice. So is the Tags filter: pick any of the tags on this brand’s prompts to narrow the list and summary cards to the prompts that carry them, or click a tag on a prompt row to filter by it. See Tags for how tags work. While either filter is on, the Unique sites and Citation events cards count the citations in the filtered list, so they drop their last-30-days caption. The You vs rivals card, the at-a-glance summary and the How your category gets answered table are the exceptions. Sources aren’t recorded per prompt, so all three are measured across all of your prompts and neither filter narrows them. While a filter is on, each of them says so.

​

Filtering by mentioned brand

If this brand tracks competitors, the Citations by recency card has a Mentioned: row in its header: one chip for your brand, marked (you) and selected by default, and one for each tracked competitor. Pick a competitor to switch the citation list to the sources cited in answers that also mentioned that competitor, on your brand’s prompts, whether or not those answers mentioned you. Click the selected chip again, or your own chip, to return to your brand. If this brand tracks more than 12 competitors, the row adds a Search competitors… box and shows at most 12 competitor chips at a time; type to narrow them. A competitor you have picked keeps its chip even when the search hides it, so you can always switch back. While a competitor is selected:

• The summary cards at the top of the page, and the at-a-glance summary, stay on your brand.
• The How your category gets answered table narrows to the domains cited alongside that competitor, with the same caveat: co-occurrence, not proof that a domain informed what the AI said about them. Only the rows change. Each row’s you · rivals counts still count every tracked competitor, not only the one you picked.
• The Branded / Non-branded and Tags filters still narrow the citation list.

If this brand’s competitors can’t be loaded, a note in the card header says the filter is unavailable, rather than the row silently disappearing. Reload the page to try again.

​

Own vs competitor vs third-party sources

Every cited domain is classified by what kind of source it is, shown as a coloured pill next to the domain name: A fifth category, Other, catches domains that don’t cleanly fit the four above — which is where other companies’ own websites land, since they have no publisher type. Each type implies a different way to earn a citation there: owned means better content on your own pages, editorial means PR coverage, UGC means genuine reviews, reference means accurate entity data on the sites AI treats as authoritative.

​

How your category gets answered

Below the citations list sits a second, distinct table: How your category gets answered. This is intelligence, not a to-do list: the table lists the domains AI platforms pulled from when answering your prompts in your latest scan, ordered by how often they were cited. It deliberately includes your own site and other companies’ sites — knowing that a competitor’s own domain anchors your category’s answers is worth having, even though you can’t act on it. Each row shows the domain, its source type, a Reach badge, and a you · rivals co-occurrence bar:

• Reach is whether presence there can be pursued at all: Your site (fix your own pages), Earnable (independent publishers, communities, directories and marketplaces — coverage can genuinely be earned), Company site (another company’s own website — one of your tracked rivals or any other business, such as a brand, retailer or shop; competitive intelligence only), Cite-only (government, academic and standards bodies — cite them, don’t pitch them), or Platform link (a search engine, AI-platform redirect or file-hosting link — not a publisher you can approach). Click or tap a badge to see what it means. A row showing a dash instead of a badge hasn’t been classified yet.
• The you · rivals counts are co-occurrence. You is how many answers that mentioned your brand cited that domain. Rivals counts competitor-mention–answer pairs — one answer naming three tracked competitors contributes three, so it scales with how many competitors you track. AI platforms return one citation list per answer, so neither count means your content (or theirs) has been retrieved from that domain.

​

At a glance

Directly under the four summary cards, Your category’s sources at a glance sums up the same source data as the How your category gets answered table. It has four parts: a short headline, your share trend, the source mix, and a What changed list. On a phone they stack into one column. Open the domains table at the foot of the card jumps down to the full table. Nothing on the page narrows this summary — not the Branded / Non-branded or Tags filters, and not the Mentioned: chips. Whenever one of them is on, a line under the card’s title says the summary is measured across all of your prompts and all of your tracked rivals. The card doesn’t appear until a scan has produced source data, and if your latest scan’s sources can’t be loaded, it shows a short message in place of the summary.

​

The headline

Up to three sentences sum up the card. The source-mix percentages are the same as the By reach legend, and your share is the You vs rivals card’s value. The earlier figure your share is compared with is your previous few scans taken together, so it won’t match any single point on the chart.

• Your source mix — in your latest scan, what share of source co-occurrences were on company sites, on earnable sources and on your own site. These are the same percentages as the By reach legend.
• Your share — your share of source co-occurrences (the You vs rivals card), compared with your previous few scans taken together. It only calls your share up or down, with an arrow alongside, when the move is bigger than the swing a share normally shows from one scan to the next at that volume. A smaller move reads in line with your previous scans. When there’s no earlier scan to compare with, it simply states your share. If the latest scan has too few co-occurrences to compare, it says so instead. If you started tracking a rival during the period being compared, it gives the earlier figure without calling a direction, because adding a rival lowers your share by itself. It does the same when the earlier scans had too few co-occurrences between them to read much into, and when fewer AI platforms answered your latest scan than the one before. If you don’t track any rivals yet, it explains that your share reads 100% only because there’s nothing to compare it with.
• What else changed — when you started tracking a rival, or the number of prompts you track changed noticeably, during that period, a third sentence says so. It also says when fewer AI platforms answered your latest scan than the one before. Any of these can move your share with no change in how AI answers. Swapping prompts for the same number of others isn’t picked up here, but What changed below flags it.

The info icon next to the headline explains how co-occurrences are counted.

​

Your share vs rivals, per scan

A line chart of your share of source co-occurrences, one point per scan. It’s calculated exactly like the You vs rivals card, whose value is always the chart’s latest point. Your tracked rivals make up the rest of each scan, so there’s no separate rivals line. Scans sit on a time axis, so uneven gaps between scans show as uneven gaps on the chart. The line above the chart gives how many scans it covers and their date range.

• Hollow points mark scans to read with caution: ones with too few co-occurrences to read much into, or where fewer AI platforms answered than in the scans around it.
• Dashed markers show where you started tracking another rival. Each one sits on the first scan that finished after you added the rival, which isn’t always the day you added it.
• Dotted markers show where the number of prompts changed noticeably from the scan before. Swapping prompts for the same number of others isn’t marked.

A key under the chart explains whichever of these appear. Hover over or tap a point for its details: the date, your share and your rivals’, the co-occurrences behind them, how many answers, prompts and AI platforms the scan covered, and any rival or prompt-set change since the scan before. Rivals you’ve stopped tracking aren’t recorded, so removing one leaves no marker. Early on, the chart says what it can’t show yet. With a single scan, it shows that scan’s share as your baseline and says your line starts with your next scan. With only a few scans, a note above the chart reminds you that your share often moves several points from one scan to the next, so small moves on the line aren’t a direction yet. If you don’t track any rivals yet, it says your share reads 100% with nothing to chart, and links to Add competitors for this brand. If none of your tracked rivals appeared alongside these sources, it says so, since your share is then 100% on every scan.

​

Where AI answers get their sources

A single bar splits your latest scan’s source co-occurrences into groups. The legend gives each group’s percentage and how many domains are in it, and the line above the bar gives the total co-occurrences, the number of domains and the date of the scan. Two tabs switch the grouping:

• By reach (the default) groups sources by Reach. Your site, Earnable and Company sites are always listed, even at 0%, because an absence is worth seeing. Cite-only, Platform links and Not yet classified appear only when your scan has some. Company sites include every business’s own website, not only the rivals you track, and each domain’s Reach reflects its latest assessment.
• By type groups the same co-occurrences by source type: Owned, Editorial, UGC, Reference and Other, always all five. Other companies’ websites have no publisher type, so they count as Other. When Other dominates the bar, a note under it points you to By reach, which shows them separately.

The bar, the legend and the headline share one calculation, so their percentages always agree and add up to 100%. A group that’s present but too small to round up to 1% shows as <1% in the legend and “under 1%” in the headline. The legend is read-only, and switching tabs here doesn’t change the table’s type filter. The info icon explains each Reach group.

​

What changed

Which sources gained or lost ground between your most recent scans and the same number of scans just before them. The caption under the heading names the exact scans and dates being compared. Movement is measured as each source’s share of all co-occurrences in each window, not as raw counts, so extra scans or re-runs don’t read as movement. Three tabs, each with a count:

• Rising — sources whose share went up.
• Falling — sources whose share went down. A source with no co-occurrences at all in the recent window is tagged Dropped out.
• New — sources that first appeared in your scans during the recent window. This tab appears once you have a few scans of history; until then, the panel says when new sources will start to show.

The panel opens on Rising, or on Falling when nothing rose. To make the list, a source has to clear every one of these tests: it’s present in enough of the scans it’s measured on; it has at least a few co-occurrences; and its share moves by both a meaningful number of percentage points and a meaningful proportion of its earlier share, even after allowing for how much busier or quieter the recent scans were. These tests keep the list short. They aren’t a test of statistical significance, and a source that jumps often settles back, so read the list as what changed, not as a trend. The How movers are picked info icon lists the exact thresholds. Each row shows the domain with its Reach tag, a bar running from its earlier share (hollow dot) to its current share (filled dot), both percentages, and the change in points. New sources show their current share and the date they first appeared. Direction is shown by an arrow and a sign, not by colour, and every bar across the tabs shares one scale, noted under the list. The first few rows show straight away. Show all expands the rest, and a very long list shows its largest movers and says how many there are in total. Notes above the tabs flag anything that makes the comparison less clean:

• Your prompt set changed between the two windows — some of the movement then reflects different questions, not a changed market.
• You started tracking a rival between them — sources cited alongside that rival gain share for that reason alone.
• Only one scan on each side — early on, the panel compares your latest scan with the one before, and says to expect some noise.
• A thin earlier window — when the earlier scans recorded too few co-occurrences for reliable shares, those shares are flagged as rough.

Before your second scan there’s nothing to compare, and the panel says so; it says the same when either window has no co-occurrences at all. If the comparison can’t be loaded, the panel says that too — the rest of the page is unaffected.

​

Acting on what you see

Where a domain is genuinely Earnable, coverage there improves the odds future answers cite you — match the approach to the type: Editorial means pitching press coverage; UGC means genuine community participation; Reference means correcting your entity’s listing. Your site rows are fully in your control: make the pages answer the questions your prompts ask. Company site, Cite-only and Platform link rows are context, not tasks. The same goes for What changed: a company site rising is worth knowing about, not something to chase. AI answers vary run to run, so judge progress over several scans, never a single one — the signals worth watching are your inclusion rate on the prompts that cite a domain, whether your own pages start appearing in the Citations list, and the direction of your share trend once it has a few scans behind it. No placement guarantees a mention.

​

Related pages