The short version.
- AI systems may retrieve publishers, directories, reviews, forums and other outside sources.
- Original studies show a relationship between earned mentions and AI visibility, but correlation is not causation.
- A relevant mention without a backlink can still help establish category context.
- The goal is accurate consideration in credible sources, not manufactured mentions.
Not every mention carries the same weight.
Select a source type to see what makes it useful.
Strong when the page directly compares the category and stays updated.
Illustrative score based on relevance, credibility, freshness and contactability.A recommendation needs evidence outside the sales page
A brand website is the best place to publish official facts, positioning, specifications and proof. It is also a source controlled by the company making the claim. When a buyer asks for the best option, an answer is more useful when it can compare that claim with independent evidence.
That evidence may come from a listicle, a review profile, a trade publication, a directory, a retailer, a forum or a video transcript. Different AI systems use different retrieval methods, so the exact source mix changes by platform and prompt.
This is not an argument that every outside mention is automatically trustworthy. It is a reason to understand which pages repeatedly shape the comparison and whether they represent the category accurately.
What the current studies suggest
Muck Rack's May 2026 update reported that earned media accounted for 84 percent of the AI citations in its dataset. Its research also shows that source behavior differs by model and that prompt wording can change which sources appear.
Ahrefs studied 75,000 brands and found branded web mentions had a strong correlation with visibility across ChatGPT, Google AI Mode and AI Overviews. The reported correlations ranged from 0.66 to 0.71. Ahrefs is careful to state that correlation does not prove causation, and brands should be equally careful.
Together, the studies support a reasonable working hypothesis: accurate earned coverage can contribute to the evidence environment around a brand. They do not provide a recipe that guarantees an AI recommendation.
- Treat the studies as directional evidence, not a ranking formula.
- Track the pages that repeatedly appear for commercially relevant questions.
- Separate a brand mention from a page citation because they are not the same outcome.
- Review results across multiple prompts and platforms before drawing a conclusion.
A mention can matter without a backlink
Traditional SEO reporting often centers the link. AI visibility adds another useful unit: the unlinked mention. A page can describe a brand, place it in a category and compare it with alternatives even when it does not link to the company website.
Semrush's 2026 ghost-citation study found that many citations did not produce a brand mention. That is a reminder to track both sides. Publishers may care about being cited. Consumer brands usually care more about whether the brand itself appears accurately in the answer.
A measurement sheet should therefore record the page URL, the brand mention, the outbound link and the AI citation as separate fields. One can change while the others remain flat.
The responsible play is relevance, not volume
The wrong conclusion is to chase mentions everywhere. Google explicitly warns against inauthentic mentions and other attempts to manipulate its systems. A placement campaign should begin with pages where the brand genuinely belongs and where inclusion would make the resource more complete for the reader.
That is why opportunity qualification matters. A small trade guide with the right audience may be more useful than a large unrelated publication. The page, category, audience, editorial standard and contactability all matter.
A relevant page can still be a poor opportunity if it is abandoned, inaccurate or impossible to update. Qualification protects the brand and prevents outreach volume from becoming a vanity metric.
Find the source before you find the contact
Begin with the page, not a giant media list. Search the questions a buyer asks, inspect the pages appearing in search and AI answers, and note where competitors are already included. Then decide whether the client genuinely fits.
Only after the page passes that test should contact research begin. The useful person may be the named writer, a commerce editor, a category manager or the owner of a small directory. Contacting the wrong person quickly is not progress.
A strong pitch references the exact page, explains the missing value and supplies evidence the editor can check. It does not lead with a backlink request or pretend that inclusion is owed.
Report the campaign without pretending you control the editor
The agency can guarantee the research, qualification, outreach, follow-up and reporting described in the agreement. It cannot guarantee that an independent publisher will update a page or that an AI system will use it later.
Report the full funnel: opportunities reviewed, contacts verified, pitches sent, replies, active conversations, declines, updates and published placements. Add AI mention and citation tracking as a separate observation layer.
That level of detail makes the work easier to judge. It also keeps a slow editorial process from being hidden behind a single placement count.