Find third-party pages worth approaching from your AI citation data

To find third-party pages worth approaching, start from the pages AI answers already cite, then keep only those that genuinely cover your category, that do not yet mention your brand in their saved text, and that a human editor could reasonably update. Rank the shortlist by how many completed answers and prompts cite each page, review each page by hand, and record every step. Citation counts tell you which pages AI engines link most often. They do not tell you whether a publisher will respond, or whether a placement will change what an engine says about you.

In short

• Start from cited pages, not from a media list. The pages AI answers link are evidence you can inspect; a generic outreach list is not.
• Filter in this order: relevance, then the brand-match check, then editorial fit, then citation count. Count alone is a poor first filter.
• Score each candidate with a review sheet and write down the reason for every "approach", "hold" or "skip".
• Outreach happens outside your monitoring tool. Keep a log so you can compare citations before and after a reported placement.
• A before-and-after change in citations is an observation, not proof that your outreach caused it.

What is an earned source, and why look at cited pages?

An earned source here means a third-party web page (not your site, not a competitor's) that AI answers cite and that you could try to appear on through editorial means. A citation is a link in an AI answer pointing at a specific URL as a source; you can read more in the citation glossary entry.

Looking at cited pages narrows a huge outreach universe to pages that engines have actually linked for the questions you monitor. That is a useful starting filter. It is still only a filter. A page can be cited for reasons unrelated to how it treats your category, and being on a cited page does not guarantee that an engine will name you next time.

Enlarge image

How do you get a list of cited third-party pages?

Build the list from your saved citation data for a fixed window. In DiscoveredBy, the Earned sources screen has a Discover cited pages view: pick a 7, 28 or 90 day window, and it lists pages cited in completed answers in that window. It leaves out your own domain and your active tracked competitors' domains, and it orders pages by the number of distinct completed answers citing them, then by distinct prompts.

Three properties of that list matter for how you read it.

• Repeated links in one answer count once. A page cited three times in a single answer adds one to its count, so the count reflects reach across answers, not link volume.
• Counts pool channels and countries within whatever the filter bar has narrowed to. Narrow by engine, country, persona or language if you want a cleaner picture for one market.
• The counts are not an authority score. The docs say plainly that they are a way to find observed sources, not a likelihood of placement, and not a claim that the page is missing your brand.

If you know of a relevant page AI has not cited yet, you can add it by hand. A hand-added page carries no citation evidence, and nothing is fetched or scored when you add it, so treat it as a different class of prospect in your sheet.

Enlarge image

Which pages are worth your time first?

The pages worth your time first are the ones that are on-topic, plausibly open to an addition, and not already covering you. Citation count breaks ties among those; it should not decide the shortlist on its own.

Work through five checks for each candidate.

1. Category relevance. Does the page treat the same problem your product solves? A page about a neighbouring topic that happens to be cited for one prompt is a weak prospect.
2. Brand-match status. Does the saved text of the page name your brand? DiscoveredBy's Mentions view checks literal names and aliases in saved, extracted text of cited third-party pages. It can return a match, no match, or unknown (for example a failed, blocked or oversized scrape). Unknown is not the same as absent.
3. Competitor presence. Does the page name competitors but not you? The Mentions view can filter for a competitor match without your brand. That is a fact about the saved text, not a reason the page is unfair to you.
4. Editorial fit. Is there a real reason for the author to add or update something: a gap in coverage, an outdated section, a missing use case? "Please mention us" is not a reason. A concrete correction or addition is.
5. Freshness of the evidence. Check the fetch date on the saved text. Older successful text remains analysable, so a page may have changed since it was read.

What can the saved evidence tell you, and what can it not?

It can tell you which pages were cited, how often, by which channels, and what the saved text of the page contained when it was read. It cannot tell you why an engine chose the page, whether the page influenced the answer, or whether the live page says the same thing today.

When you save a discovered page as an opportunity, the record keeps the URL, title, domain, discovery dates, distinct answer and prompt counts, and up to five citation examples with collection date, recorded channel and a saved snippet of citation context. That evidence stays attached and is not refreshed by later collection. A saved opportunity's counts cover every completed answer in its window, not only the ones your filters selected. A saved snippet does not verify a brand mention on the live page.

Two consequences follow. First, re-check the live page before you write to anyone. Second, remember that stated reasons, citations and mentions in an answer are observations of that answer. They do not reveal what internally led an engine to produce it. If a page is cited alongside a competitor, you have learned that both appeared in that answer's sources, not that the page caused the competitor's placement.

For related reading on the reverse case, where a publisher covers your category and leaves you out, see A cited publisher covers the category but omits your brand. This post is about ordering a whole list of candidates.

Enlarge image

The prospect-review sheet

Use one row per page. Copy this table into a spreadsheet, or use the text template below it.

PROSPECT REVIEW
Page URL:
Window / filters used:
Answers citing (distinct):        Prompts citing (distinct):
Page type:
Covers our category?  yes / partly / no
Brand match: match / no match / unknown   (text fetched on: )
Competitors named on page:
Live-page check (date, what I saw):
Editorial angle (specific addition or fix):
Decision: approach / hold / skip
Reason (one sentence):
Owner:            Status:            Follow-up date:
Outreach notes (who, when, channel, reply):
Placement reported on (date):   Evidence:

How to make the decision column consistent

Use the same rules for every row, so two reviewers reach the same answer.

The "Approach" rule requires an angle in the sheet. If the angle cell is empty, the decision is not "approach".

How do you run and record outreach?

Do the outreach in your own email or PR tooling, and record it in the sheet or in the opportunity record. The Earned sources workflow is a place to review sources and coordinate the work. Per its documentation, it does not send messages, find contacts, buy placements, publish content or connect to a placement partner.

Inside DiscoveredBy, an opportunity has an assignee, notes (up to 4,000 characters), a follow-up date and one of five statuses: New, Planned, Outreach in progress, Placement reported and Dismissed. Owners and editors can save and update opportunities; assignees must be active owners or editors, and viewers can read but not edit. Follow-up dates appear on the record; they do not schedule messages or notifications.

Keep the outreach itself simple and honest:

• Lead with the specific gap or correction, not a request for a link.
• Give the editor something they can verify: a source, a dated fact, a current product page.
• Do not promise coverage, and do not buy placements you would have to hide.
• Log each contact and reply in the notes, including refusals.

When you mark a status as Placement reported, the tool requires a date no later than today. That is a claim your team enters, not independent verification, so add your evidence to the notes.

How do you check what happened after a placement?

Compare the page's citation rate in equal periods before and after the reported placement. The Observed citations panel does this for a saved page using 7, 28 or 90 day periods, and it excludes the placement day because answers that day may predate the change.

The measure is: page citation rate = completed answers citing the saved page (or a recorded address variant of it) / completed answers × 100, calculated separately for each recorded provider, platform, surface, collection method and model. The comparison is unavailable until the full after window has elapsed. A period with between 1 and 29 completed answers is marked provisional, and a period with none is unavailable rather than zero.

Read it cautiously. The docs state that a higher rate does not establish that outreach caused it, that the source now mentions your brand, or that revenue changed. Prompt and country mixes may differ between periods, and the panel uses live saved observations that late collection can revise. For a broader view of how a domain's presence moves, Source comparison compares equal periods of cited sources, and Source-mention history lets you save dated snapshots of what cited pages said about your brand.

Worked example: Quillstone builds a shortlist

Illustrative example: Quillstone and its competitors are fictional, and the numbers are made up to show the method.

Quillstone sells document-review software to mid-sized legal and compliance teams. Its content lead reviews the last 28 days of cited third-party pages, with all engines pooled, and copies the top candidates into the sheet.

The lead makes these calls:

• lawtechreview.example: Approach. It is on-topic, names two competitors and not Quillstone. She opens the live page and finds the review of workflow features is two years old. Her angle is a factual update on Quillstone's current approval-routing feature, with a link to its documentation page. She does not ask for a ranking position.
• compliancedaily.example: Approach, lower priority. No brand match and no competitors either. The angle is a missing section on review audit trails, offered as a contributed paragraph.
• legalops-forum.example: Skip. Quillstone already appears in the saved text. She checks the snippet; it is accurate.
• lawlibrary.example: Hold. The match status is unknown, so she opens the page. It turns out to be a students' reading list with no editorial contact, so she changes the decision to skip and writes the reason.

Two rows go to "Planned", assigned to a teammate with a follow-up date. She emails the editors from her own mailbox. On August 10 the first publisher reports the update is live; she sets Placement reported on that date, saves a screenshot in the notes, and waits for the after window.

For a seven-day comparison, the panel uses August 3–9 and August 11–17, and becomes available on August 18. For one channel it shows 6 of 40 completed answers citing the page before (15.0%) and 9 of 42 after (21.4%), a change of +6.4 percentage points. Both periods have at least 30 completed answers, so neither is provisional.

What she writes in the report: "Citations of this page rose from 15.0% to 21.4% of completed answers in one channel after the reported update." What she does not write: that the outreach caused the rise, or that the page now helps Quillstone's mention rate. She has not checked either.

Common mistakes

• Sorting by count and emailing the top ten. The most-cited page may be irrelevant or already accurate about you.
• Treating unknown as no. A failed or oversized scrape leaves the brand-match status unknown; it is not evidence of absence.
• Relying on old saved text. Check the fetch date and open the live page.
• Asking for a mention instead of offering a correction. A verifiable improvement gives an editor something to act on; a request for exposure does not. Whether a specific publisher will respond is something no citation data can predict.
• Reading a rise as proof. A comparison across two periods with different prompt mixes cannot isolate your outreach.
• Working a page with no owner. Unowned rows stall.
• Forgetting the competitors and your own domain are excluded from discovery. Competitor pages are measured elsewhere, on the Competitors screen.

If you already have a specific gap you want to close on your own site instead, see Find the citation gaps that deserve your next content update, and to choose between updating a page, writing a new one or earning a mention, see Update a page, write a new article, or earn a mention.

Frequently asked questions

Should I approach every page AI cites?

No. Many cited pages are off-topic for your category, already accurate about you, or closed to outside input. A short list with a concrete angle for each page is easier to act on and review than a long list of generic requests. The sheet's decision column exists to make that cut explicit.

Does getting mentioned on a cited page make AI name my brand?

Not reliably, and no source in this workflow claims it does. Engines assemble answers from many inputs. Track the page's citation rate and your own brand mentions over time, and treat any movement as something to investigate, not as proof of cause.

Can the tool find contact details or send outreach for me?

No. The Earned sources workflow does not send messages, find contacts, buy placements or publish content. You use your own email or PR tools and record the work in the opportunity's notes and status.

How long should I wait before judging a placement?

Until the full after window has elapsed. With a seven-day comparison and a placement on August 10, the after window is August 11–17 and the panel becomes available on August 18. A period with between 1 and 29 completed answers is marked provisional, so choose a longer window if your volume is low.

Can I save a competitor's page or my own?

Neither can be saved. Your own registrable domain and the domain of an active tracked competitor are excluded from discovery and refused when you add a page by hand. Earned sources are third-party pages by definition.

Are review sites and forums worth approaching?

They can be candidates, but the approach differs. A forum thread or review listing is usually not something an editor can revise for you. Mark the page type in the sheet and decide whether a legitimate route exists; see Evaluate a cited "best tools" list without copying its rankings for editorial lists.

Next step

Open your citation data for the last 28 days, list the top ten third-party pages, and fill in the sheet for each before you write a single email. DiscoveredBy's Earned sources workflow keeps the saved evidence, assignments, statuses and post-placement comparison in one place, and the wider picture is in Sources. Sign in or start at app.discoveredby.ai to build your first shortlist.