Find the citation gaps that deserve your next content update

A citation gap is a question your buyers ask where an AI answer links to someone else's page and does not link to yours. Not every gap deserves work. To find the ones that do, start from the buyer question, read the page the engine cited, decide who owns it and what your side is missing, then choose one of three actions: update a page you already have, create a page you do not have, or investigate a source you do not control. Score each gap on buyer value, evidence, reach and effort, and work the top of the list. The worksheet below does exactly that, and every suggested change is a hypothesis to test, not a promise of a citation.

In short

• Begin with a buyer question in your tracked prompts, not a page you already wanted to rewrite.
• Read the cited page first. The gap is the specific thing it offers that you do not.
• The owner of the cited source decides which of three actions fits.
• A monitoring tool ranks gaps by evidence; only your team can rank them by buyer value.
• Record the change date, and treat any later movement as an observation, not proof.

What is a citation gap?

A citation gap is the difference between what the pages an AI engine cites for a prompt contain and what your own page offers for the same question. It is the specific, missing piece, such as a comparison table, an answer to a follow-up question or a figure your page never states. It is not a general feeling that a competitor "ranks better".

Three neighbouring terms are worth keeping apart. A citation is a link an engine puts in its own answer, pointing at a URL as a source. A mention is your brand named in the answer, whether or not it is linked. A retrieval is a page the engine read but did not link. This post is about citations. If your page appears to have been read but is not linked, that is a different problem, covered in Your page was retrieved but not cited: what to check next. For the shared definitions, see Key terms and the glossary entry for citation gap.

Where do you find candidate gaps?

Look in three places, in this order: the Citation gaps queue, the Citations screen filtered to external sources, and Source comparison.

Citation gaps queue. In DiscoveredBy, the Citation gaps screen (available depending on your plan) takes the prompts where you are not winning and lists specific proposed changes, ranked by an impact score. A prompt only becomes a candidate when, over the trailing window, you appear in under 30% of that prompt's runs and at least one citation points at a domain that is not yours. For each candidate, the product reads the five most-cited competing URLs, finds your best matching page, and records each missing signal as evidence. If no page of yours matches closely enough, it recommends creating one instead. A prompt you already win most of the time never becomes a gap.

Two limits matter. The impact score is a relative order within your project, not a percentage of anything. And a run keeps only a plan-derived number of opportunities, dropping the lowest-impact first, so an absent gap does not prove there is none (see plans and limits).

Citations screen. The Citations screen can filter to External, meaning every citation that is not yours, competitors included. It is built from the newest 500 citations matching your filters, and the screen says so when the window holds more.

Source comparison. Source comparison compares two equal, adjacent periods (7, 28 or 90 days) and shows, for each cited domain, the share of completed outputs citing it, and the change in percentage points. Use it to tell a source that has always been there from one that has just appeared or just vanished.

Enlarge image

How do you decide what to do about one gap?

Read the cited page, then ask who owns it and whether you already have a page that should answer the question. Those two answers pick the action.

Step 1: pick the buyer question. Start from a tracked prompt written the way a buyer would ask it; a gap on a non-buying question is not worth fixing. See How to choose AI tracking prompts that reflect buying decisions.

Step 2: read the cited page. Open the top cited URLs for that prompt and read them. Note what they contain that yours does not: a table, a definition, a worked figure, a step list, a stated limit. Write it as a plain sentence. "Has a feature table comparing three products" is useful; "is better" is not.

Step 3: classify the source's owner. The cited page is either a competitor's own page, a third party's page (a publisher, a review site, a community thread), or, less often, one of yours that is cited but not enough. In DiscoveredBy, ownership is decided from the domain: a citation is yours when its registrable domain is your project's domain, and nothing about the page's content is checked. A copy of your article republished on another site therefore counts as that site's citation.

Step 4: pick the action.

Here the action is a triage label that lets you compare unlike gaps on one sheet; weighing the three in depth is a later topic.

Step 5: score it. Score the four factors in the next section and add them up.

Step 6: record it. Log the prompt, action, date and review window. Accepting a gap in DiscoveredBy also creates a candidate topic on the Articles screen with the diagnosis attached; nothing is written until you pick it there.

How do you use source comparison to check a gap is real?

Use it to check whether the cited source is persistent or new, and whether its citation rate is large enough to matter. It cannot tell you why the source is cited.

With your domain and the competitor or publisher domain in view, compare two equal periods on one recorded collection channel. Each row shows how many completed outputs cited the domain in each period, each count as a percentage of that period's completed outputs, and the difference in percentage points. For example, 2 of 10 outputs followed by 6 of 20 outputs is 20% followed by 30%, a change of +10 percentage points.

Caveats: a period with fewer than 30 completed outputs is marked provisional (a sample-size cue, not a significance test). A zero means the collected records had no citation for that domain, not that the engine never cites it. Prompts and countries are pooled within each channel and can change between periods, so it is not a fixed-cohort experiment. It does not establish source authority or why a source moved.

It also reports your owned-source share: your own-domain appearances divided by all domain appearances, where an appearance is one domain cited in one completed output. It is a useful backdrop, but it does not replace reading the page.

The prioritization worksheet

Copy this into a spreadsheet or a ticket. One row per buyer question, four scores, one action.

CITATION GAP WORKSHEET

Buyer question (the tracked prompt):
Window and filters used (dates, engine, channel, country):
Own presence in that window (appearances / completed outputs):
Top cited source(s) that are not yours (URL, owner type):
What that source has that we lack (one plain sentence):
Our best matching page (URL, or "none"):
Action: UPDATE owned page | CREATE new page | INVESTIGATE third-party source | NOT A GAP
Hypothesis (what we will change and what we expect to observe):

Scores, each 1 to 3:
  Buyer value  (1 = side question, 3 = decides the shortlist):
  Evidence     (1 = one thin observation, 3 = repeated, specific, readable):
  Reach        (1 = one prompt, one engine, 3 = several prompts or engines):
  Effort       (1 = an afternoon, 3 = new page, review or outreach):

PRIORITY = Buyer value + Evidence + Reach - Effort

Owner / due date:
Date changed:
Review date and window (same prompts, same channel):
Result (improved | no change | insufficient evidence):

Use the scores like this.

The 1 to 3 scoring and the priority formula are our own method for structuring your team's judgement. They are not a DiscoveredBy feature, and the product does not compute them. The impact score inside DiscoveredBy weighs citation share, how far under you are, engine spread, evidence found and how substantial the competing pages are; it knows nothing about which questions decide your sales.

Enlarge image

A worked example

Quillstone sells document-review software to mid-sized legal and compliance teams. Its team tracks four buyer questions over a 28-day window. Each was collected on 20 completed outputs.

The arithmetic: row one is 3 + 3 + 2 - 1 = 7, row two is 3 + 2 + 3 - 2 = 6, row three is 2 + 2 + 2 - 3 = 3. The fourth prompt is at 45%, above the 30% threshold the product uses, so it never becomes a gap in the queue. The team keeps it under watch.

What the team does:

1. Row one, update. They read the Brieflane page and find a table comparing review turnaround by document type. Quillstone's product page has no comparable table. The hypothesis: adding a factual, sourced table of Quillstone's own review features to that page may make it a stronger candidate for this question. They record the edit date and the review window.
2. Row two, investigate. They cannot edit a publisher's roundup. They read it, check what it says about the category and whether Quillstone is mentioned, and note that its listed criteria fit the product. Whether to approach the publisher goes to the outreach owner, since contact and outreach happen outside the tool.
3. Row three, create. No Quillstone page matches. It goes on the list as a new-page candidate. The brief for it is the next step, and Turn an AI citation gap into a content brief your writer can use covers that.

They work row one, then row two's investigation, then row three. If the row-three question turned out to decide more deals than assumed, they would raise its buyer value and the order would change. The reasoning is written down and can be revised.

How do you handle a gap on a third-party source?

You cannot edit the page, so the work is to understand it first and only then decide on outreach. Start with what it covers and whether it already mentions you.

In DiscoveredBy, the Earned sources workflow lets you find cited pages that are not yours or a tracked competitor's, save them as opportunities, and record outreach and placements. It ranks them by the number of distinct completed outputs that cite them, then distinct prompts. That is a way to find observed sources; it is not a source-authority score or a likelihood of placement. The tool does not send messages, find contacts, buy placements or publish. Your team does the outreach elsewhere. The Mentions view under Sources (reached from the Domains screen) checks the saved text of third-party cited pages for literal matches of your brand and your active tracked competitors. It reads the saved text of at most 500 pages per request, and pages with missing, failed or old scrapes show as unknown or carry a fetch date. Because it matches names literally, no match does not prove the page never mentioned you, and a match on a common word may be unrelated.

Before spending effort, ask whether the source covers your category, whether including you would be accurate, and whether it appears across several prompts or only one.

Common mistakes and what this cannot tell you

• Sorting only by impact score. It is a relative order inside one project's evidence. It does not know which questions win deals.
• Treating a missing element as the cause. A competitor's page having a table and yours lacking one is an observed difference. The product records evidence of what is missing, and that is not proof of why an engine chose one page.
• Ignoring the window. A gap seen on a thin sample or one channel may not hold on another. Note both, and keep them the same when you review.
• Assuming a fix will produce a citation. In DiscoveredBy, marking a gap done sets it to implemented; once visibility for its prompts is re-checked at 7, 14 and 30 days after that, the status can move to won, no lift or monitoring. A change in visibility is an observation, not proof that your edit caused it. See Did your content update help? A practical before-and-after review.

Frequently asked questions

How many citation gaps should I work on at once?

As few as you can finish and review. One change with a clean before-and-after window teaches more than five made together, because you cannot tell which one mattered. Take the top one or two rows per cycle.

What if the cited source is a competitor page I cannot match?

Then it is a signal, not a task. Ask whether your product should answer that question at all. A gap you would not want to win should score low on buyer value or be dropped.

Should I always create a new page when no page matches?

Not always. Create when the page is truly missing, the question matters to buyers, and you have real evidence to put on it. Otherwise you may be adding a thin page.

Does a gap on one engine mean I have a gap on all of them?

No. Citations are collected per engine and channel. Check the engine and channel filters, and see the engines reference for what each surface exposes.

How often should I refresh this worksheet?

At a cadence that matches how often you can act. The queue re-reads eligible prompts each run, so a prompt left out once is not permanently excluded.

Next step

Open the Citation gaps screen for your project, copy the top five opportunities into the worksheet, and read the cited pages before you score anything. Then pick one action, log the date, and review it in the same window later. To see the queue against your own tracked prompts, sign in to DiscoveredBy. For more on the source-side tools, see the Sources feature and the Optimize feature.