Link Building Outreach: Which Placements Are Worth Pursuing For AI Citations

Rank every outreach prospect by whether AI engines already cite its kind of source for your category, and pitch only the ones that pass. Domain rating alone is the wrong filter. In Muck Rack's May 2026 analysis of more than 25 million AI-cited links, 84% came from earned sources a brand neither owns nor pays for.

This holds for B2B SaaS teams with a finite monthly pitch budget and a prospect list of any size, because the rule needs only the list and a set of citation checks.

Citation checks are snapshots, since AI answers vary between runs, so a pass is a reason to pitch and not a guarantee of a citation.

Why Link Building Outreach Needs A Ranking Rule

Outreach time is the scarce resource, and most guides never say which sites deserve it. They tell you how to find an email address, how to write a subject line and when to follow up. They skip the decision that matters most, which is who you should email at all.

Link building still works. Link building mainly improves rankings and only indirectly affects AI visibility. Nobody serious is arguing you should stop. The sharper question is which placements also show up in the answers your buyers read in ChatGPT, Claude and Gemini.

That is where link building and AI brand mentions part ways. Link building mainly improves rankings and only indirectly affects AI visibility. Brand mentions in third-party sources that AI systems already cite affect it directly. The two overlap, but they are not the same job, and a site can be excellent at the first and useless at the second.

The size of the earned slice is the reason this matters. Muck Rack's May 2026 analysis found that earned media, meaning sources a brand neither owns nor pays for, accounted for 84% of AI-cited links across more than 25 million links in 17 industries. If most of what AI engines cite is earned, then the sources you choose to pitch decide most of your outcome.

So the plan is simple. Before you write a single email, put every candidate through a short, repeatable rule. Pitch the sources that pass. Drop the rest without a debate.

What Link Building Outreach Is

Link building outreach is contacting site owners, editors or journalists to earn a link or a mention to your content.

That is the whole definition. It covers a guest post pitch, a data study sent to a reporter, a note to an editor about a resource page and a request to fix a broken link. What changes between them is the hook and the person you contact.

The ranking rule in this article sits in the middle of the process. Prospect sourcing comes first, where you build a list of candidate sites. The rule comes next, where you score and cut that list. The pitch comes last, and only for the sources that survived. Most teams run these in the wrong order. They find a contact, write the email and only then ask whether the site was worth the effort.

If you want the wider programme that outreach belongs to, read our AI visibility strategy guide. This article stays on one decision, which placements are worth pursuing.

Where AI Engines Get The Sources They Cite

For a B2B SaaS category, a useful way to think about AI-cited sources is to group them into five intent-aligned buckets that are worth targeting.

• Category definers: Sites that explain what the category is and set the vocabulary buyers use.
• Integration ecosystems: Partner directories, marketplaces and documentation hubs for the platforms your product connects to.
• Comparative research: Reviews, comparisons and analyst-style roundups that weigh products against each other.
• Developer knowledge bases: Technical references, forums and docs that practitioners search when they build.
• Durable news publications: Outlets whose articles stay indexed and quotable long after the news cycle moves on.

The selection test across all five is the same. Choose sites by topical proximity and editorial standard before you look at raw authority. A site that covers your category carefully and edits what it publishes beats a bigger site that touches your topic once a year.

There is one more filter worth adding. Prioritise sources that already appear in AI answers for your category, even when other sites have higher authority. A modest industry publication that engines already quote is a better bet than a famous domain they never mention.

This is also why digital PR sits so close to the question. Earned coverage in the right publications is what produces the third-party mentions engines pick up, and our guide to digital PR for AI citations covers that side in detail. If you want help being cited as the answer itself, that is the work behind our answer engine optimization service.

Where Candidate Sources Come From

Sourcing is not the point of this article, so this section stays short. Three routes feed the list, and each one produces a different kind of candidate.

• Competitor backlinks: Pull the sites that link to your closest rivals. Reviewing competitors' backlink profiles is a standard prospecting step, and it shows you where the category already earns links.
• Gap analysis: Compare your link profile with several competitors and keep the domains that link to them and not to you. A gap analysis is another standard prospecting step, and it turns a long profile into a short, specific list.
• AI answers for the category: Run your buyer questions through the engines and write down the sources they cite. This route is the one most outreach guides skip, and it is the one the ranking rule trusts most.

For the competitor side of the list, Semrush's backlink analysis and gap tools show which domains link to rival brands and not to yours, which gives you a candidate list worth running through the ranking rule. If you want tooling for the whole sourcing step, our roundup of AI backlink prospecting and outreach tools covers what this article does not.

One ordering point. Finding the right person on a site comes after a source passes the rule, not before. Hunting for an editor's email at a domain you will later reject is wasted time.

The Five-Test Ranking Rule For Placement Sources

This rule is our method. It is a decision tool built from selection principles, not a study result, and it should be judged on whether it saves you time and cuts bad pitches.

Each prospect is scored pass or fail on five tests. Pitch a source when it passes every test that applies to it. The fifth test applies only to sources where you already have a past placement, so a brand-new prospect is scored on the first four. A half-pass, such as a source that clears relevance and editorial standard but shows no citation evidence yet, goes on a hold list. Re-run the citation check once on a different day. If it still shows nothing, say no.

Notice what is missing. Domain rating and traffic are not tests. Prospect qualification starts with relevance, not with metrics such as DR or traffic, and a rule that opens with a number will happily pass a large site that has nothing to do with your category. Metrics can still break a tie between two sources that both pass. They never make a failing source pass.

The order matters too. Relevance and footprint risk can be judged by looking at a site for a few minutes, so run them first. The citation evidence test costs real time, because it needs repeated runs across engines, so it only gets spent on sources that already survived the cheap tests.

The rule in one sentence. Score each prospect on topical relevance, editorial standard, AI citation evidence, footprint risk and a company-named citation check for past placements, and pitch only sources that pass every test that applies.

How To Check Whether A Source Gets Cited

A generic category question undercounts placement citations. In a 57-placement test with 995 answers, AuthorityTech checked for exact-URL citation and found that company-named questions were cited 17% of the time, article-angle questions without the company name 11%, and the narrowest single-story questions 7%.

The practical lesson is that the question you choose changes the verdict. Judge a placement with three question types, company-named, article angle and narrow single story, and log them separately before you give a verdict. If you test only a broad category question, you will call good placements failures.

Here is the audit, in order.

1. Write 20 To 30 Buyer Questions: Use the competitive questions your buyers actually ask when they compare options in your category. Keep the wording natural, the way a person types it into an engine.
2. Add The Three Question Types: For every placement you are judging, include a company-named question, an article-angle question without your name, and a narrow single-story question.
3. Run Each Question Several Times: Use several AI engines and run each question five to ten times, because answers vary from one run to the next.
4. Log Cited, Mentioned Or Absent: For each placement, record whether the exact URL was cited, whether the brand or source was mentioned without a link, or whether it was absent.
5. Map Cited URLs Back To Sources: Group the cited URLs by domain and by type of site. This is where the pattern shows up, because you start to see which kinds of source engines return again and again.
6. Keep The Three Question Types Apart: Report the rates separately. A blended number hides the gap between the company-named result and the narrow one, and that gap is the information.

These snapshots are not final. The same question can return different sources tomorrow. Treat a single run as a clue, run the set again before you decide, and record the date next to every log entry in your own sheet.

For prospects you have never been placed on, you cannot run the company-named question yet. Use the category and article-angle questions to see whether the source type is cited at all, and save the company-named check for after the placement goes live.

Applying The Rule To A List Of 50 Domains

This walk-through is a hypothetical example. Every number below is made up to show how the rule behaves. None of it is data from a study or from a client.

Start with a list of 50 domains pulled from competitor backlinks, a gap analysis and the sources cited in AI answers. The tests run in order of cost, cheapest first, so that the expensive citation check only touches survivors.

In this example, relevance removes 19 domains in about an hour, because a quick look at each homepage and recent articles is enough to see that they do not cover the category. Footprint risk removes five more, mostly sites that publish near-identical pages in bulk. Editorial standard removes seven. By then the list is small enough to justify the citation check.

The citation check removes ten of the remaining 19. Those are sites that looked fine on paper but never appear in any run across several engines. Nine survive.

What do the nine look like? Mostly mid-sized industry publications, comparison and review sites, and a couple of integration or documentation hubs. Few are household names. That is normal, and it is the whole argument for this approach. The sources that engines cite for a category are often not the sources with the highest domain rating.

Order the nine by how often and how consistently they were cited across your runs, then by how closely they match your category, and only then by how easy the contact looks. If your team can pitch five a month, you know the first five. The remaining four wait for next month, after the citation log has been re-run.

If you need help assembling the list in the first place, our roundup of prospect list building tools is the place to start. And for tools that score sources the way this rule does, see the tools that suggest where to publish.

The Say-No Framework

A ranking rule is only useful if people follow it when a tempting site shows up. The say-no framework turns the five tests into drop rules a team can apply without reopening the argument.

Say no to a prospect when any of the following is true.

• Relevance fail: The site does not cover your category or a direct neighbour of it, however large its audience.
• Link network: The site belongs to a large network of sites that link to each other or sell placements in bulk.
• Thin and templated: The pages follow one template, say little, and could be swapped between clients without anyone noticing.
• Patterned anchors: The site or the arrangement expects the same anchor text every time. Patterned identical anchors are a footprint, and they are a reason to leave.
• Link exchange: The site exists to trade links. Link exchanges are the lowest-impact option, and a reciprocal arrangement adds footprint risk on top of low value.
• Double citation fail: The site passes every SEO metric but fails the citation check twice, on two separate days, across several engines.

The last rule is the one that causes arguments, so it needs a plain answer. Can a site that ranks well but is never cited still be worth a pitch? Sometimes, yes, if you are buying rankings and you know it. But that is a different goal from AI visibility, and it should be a conscious exception with a named owner, not a quiet drift. For the purpose of this rule, a source that is never cited loses to a source that is.

A documented no looks like this. One row in your list, five short fields.

• Domain: The site name.
• Test failed: Which of the five tests it failed.
• Evidence: The line or the log entry that shows the failure.
• Date: When you checked.
• Revisit: A date to look again, or "never" for networks and link exchanges.

Once a no is written down with its evidence, nobody has to rediscover it. A new team member can read the sheet, see why a famous domain was dropped, and move on. The sheet also protects you from the pressure to keep a source simply because someone already started a conversation with it.

Which Outreach Tactics Pass The Rule

The rule works on sources, but tactics decide which sources you can realistically reach. This section runs each common tactic through the rule at the level of type. It is qualitative on purpose, and none of it is a ranking backed by a study.

Guest posts on relevant sites are the strongest long-term authority signal. They pass the rule when the host site is relevant, edited and already cited. They fail it when the host accepts anything from anyone.

Outreach works best with newsworthy hooks, such as a launch or a data study. A generic "please link to my page" note gives an editor nothing to run, and a real reason to cover you gives them a story. Editorial and digital PR outreach passes when the publication is one engines already quote.

Link exchanges sit at the bottom. They are the lowest-impact option, and they tend to fail the footprint test as well.

We are deliberately not ranking broken link building against the skyscraper technique. The research behind this article has no dated evidence for whether either still works, and guessing would add noise. Run them through the five tests the way you would any other tactic. If the target passes, the tactic is fine. If it fails, the tactic cannot save it. Our guide to image link building covers one more tactic you may be weighing against the rule.

How The Ranking Changes The Pitch

Once a source passes, the pitch should change in one main way. Lead with the angle that source already gets cited for. If the engines keep quoting a publication's comparison pieces, pitch a comparison angle, not the product announcement you were going to send everywhere.

That single change does more than a better subject line. It tells the editor you read their site, and it points your content at the part of their archive that engines already trust.

Three smaller shifts follow from it.

• The contact: Pitch the person who owns that section of the site, not the general editor address.
• The offer: Give the readers of that source something they want, such as original data, a launch with a real news hook or a practitioner view they cannot get elsewhere.
• The follow-up: Follow up once or twice and then stop. A source worth pitching does not need a six-email chase, and a source that needs one probably failed a test you skipped.

Relevance beats metrics when you qualify prospects, and it beats them again when you write the pitch. A site chosen because it fits your category gives you something real to say. A site chosen because of its score gives you nothing to say.

Email templates, subject line formulas and sequence timing are out of scope here. The top results for this keyword already cover them at length. This article answers the question they skip, which is who deserves the email.

If you want a service to run the pitching, that is a reasonable choice. A link building outreach service finds the prospects, contacts the people who own the relevant sections and manages the follow-ups. The ranking rule is how you judge whether what they send you is worth having.

Measuring Placements After They Go Live

Repeat the citation log every month, on the same question set, so the numbers are comparable. For each placement, record one of three states, cited, mentioned or absent. Cut the source types that stay absent after several months, and put your pitch time into the types that engines keep returning.

Keep the question set stable. If you change the questions every month, you cannot tell whether a placement moved or your test did. Add new questions in a separate block and leave the original set alone.

Do not judge a single placement from one run. Use all three question types for each placement, and judge a source type only after you have several placements on it. One missing result is noise. A pattern across several placements is a finding.

Tools can run the log for you. Our roundup of AI visibility tools with citation tracking covers the options, and citation loss alerts tell you when a placed source stops being cited, so you can act before the next monthly review.

If you would rather have the checks run for you, our ongoing GEO and SEO reporting is built for exactly that, with the monthly log and the cut decisions included.

Common Mistakes When Ranking Placement Sources

• Ranking by DR alone: A high domain rating says nothing about whether the site covers your category or whether engines cite it, so you end up pitching famous sites that never move your AI visibility.
• Judging a placement from one generic query: A broad category question undercounts citations, so good placements get written off as failures.
• Treating one run as final: AI answers vary between runs, so a single snapshot can pass a weak source or fail a strong one.
• Pitching before the relevance test: Time spent on contacts and drafts at an irrelevant site is lost, and it is the cheapest test to run first.
• Keeping a source because it was easy to land: An easy yes is not evidence of value, and an unmeasured placement quietly eats pitch time every month.