The network signal is what independent sources say about you when you are not the one talking. Engines weight it heavily because a claim you make about yourself costs nothing to produce, while a mention that survives someone else's editorial judgment costs something, and cost is one of the few honest proxies a machine has for separating a real reputation from a wish. Build it with genuine third-party references, press, podcasts, respected lists, and peer citations, then anchor each one to your identity with structured markup so the system can attribute it to you correctly.
Step three is the one most people underbuild, and it is the one that carries the most weight. If you fix only one part of your PEO, fix this one.
Why self-claims are cheap
Anyone can declare themselves the best on their own website. Engines know this, so they weight self-description lightly. It is the least expensive signal to fake, and therefore the least trusted. Your carefully worded homepage is table stakes, not evidence, and a model that has ingested millions of homepages making similar claims about their owners has already learned to discount the genre.
A reference from an independent source is different in kind. A journalist quoting you, a podcast hosting you, a respected list naming you, a peer citing your work, each is a vote the engine did not have to take your word for. That is why third-party regard moves the needle when self-promotion does not, and it is why the mechanics of how that regard gets recorded, indexed, and cited matter more than the polish of your own copy. The mechanism behind that gap is covered in more depth in Mention, Citation, Recommendation, which separates the three actions engines take with your name.
Stop asking how to say you are the best. Start asking how to get respected others to say it for you. That single shift is the heart of the network signal.
The forms that count
Not every mention carries the same weight, and it helps to know which forms of third-party reference actually move a machine's confidence in you, rather than just feeling good to receive.
- Press mentions: being quoted as an expert source in publications the engine already trusts.
- Podcasts and interviews: your name and position, carried by someone else's platform, often with a transcript a crawler can index.
- Respected lists and directories: inclusion where buyers and engines look for the shortlist.
- Peer citations: other practitioners referencing your work, the strongest signal of genuine standing, because peers have the least incentive to inflate a rival's reputation.
How to earn them, honestly
The network signal is earned, not bought. Pitch yourself as a genuine expert source to journalists working your beat. Say yes to relevant podcasts and bring a real position, not platitudes. Get on the lists your niche respects by being genuinely list-worthy. Publish work good enough that peers want to cite it. Set a concrete target, a handful of real references in a defined window, and pursue them deliberately. Treat the underlying credibility markers, not just the mentions themselves, as the actual asset, since EEAT as Data explains why engines increasingly treat experience and expertise signals as structured inputs rather than vibes.
The mechanics: why corroboration outweighs assertion inside a model
To understand why the network signal carries so much weight, it helps to think about what a language model actually does with text at scale. During training, a claim that appears on one domain only is a single data point, tied to one author's incentives and one publisher's editorial standards. The same claim, or a materially similar one, appearing across several independent domains you do not control is a different kind of evidence entirely. Independent publishers rarely coordinate with each other, so when they converge on describing you the same way, that convergence functions as redundancy, and redundancy is one of the cheapest, most reliable trust proxies any system has, human or machine.
Retrieval systems behave similarly at query time, even though the mechanism is different from training. A domain that other domains link to and cite tends to rank higher in the search index that a retrieval-augmented assistant draws from, and it tends to get crawled more frequently, which keeps its information fresher in whatever cache the assistant consults. So a mention is not just a mention. It is a small deposit into two separate systems: the statistical patterns a model absorbed during training, and the live index a retrieval layer queries when someone asks about you today. For a deeper walkthrough of how those two systems differ and where each one matters, see Training Data vs Retrieval.
A worked example, two strategists and one query
Consider two hypothetical brand strategists, Jane Okafor and Tom Bramble, both credible, both with a clean personal site, both with a Person schema block declaring their expertise. Jane has stopped there. Tom has, in addition, three legitimate references built over the past year: a transcribed podcast appearance on a respected marketing show, a quote in a trade newsletter about positioning for fintech founders, and a citation from a peer's blog post that names him directly when discussing repositioning case studies.
Now imagine a founder asks an assistant who could help reposition a fintech brand. Both strategists' own sites describe them as exactly the right fit, in almost identical language, because that is how personal sites tend to read. But Tom's name co-occurs with the relevant keywords across four independent sources instead of one, and each of those sources is a domain the retrieval layer already treats as reasonably trustworthy. The assistant is not running a popularity contest, but the statistical and retrieval patterns both lean toward the name that shows up corroborated rather than merely claimed. Even when Tom is not proposed unprompted, if the founder names both strategists and asks the assistant to compare their credibility, the corroborated references give the model something concrete to point to, while Jane's site gives it only her own word. The logic an assistant applies when comparing two names head to head is explored further in How AI Decides Who to Recommend.
Disambiguation: why a reference has to name you specifically
A third-party reference only helps if the system can confidently tie it to you and not to someone else who shares your name. This is where the network signal quietly depends on good disambiguation practice. A mention that reads simply "Jane, a brand strategist" is far weaker than one that reads "Jane Okafor, brand strategist and founder of a fintech-focused consultancy in Lagos," because the second version gives the system distinguishing detail it can cross-reference against your other profiles. If your name is common, this matters even more, and the specific failure mode of overlapping identities is covered in Name Collision in PEO.
The other half of disambiguation happens on your own site, not on the third-party one. When you list the profiles and platforms that carry your references inside a sameAs property in your Person schema, you give the system an explicit bridge between the external mention and your canonical identity, rather than leaving it to infer the connection from context alone. The mechanics of that property are laid out in sameAs, the Most Underrated Markup, and the full schema pattern it belongs to is in the Person Schema JSON-LD Guide. Do both, earn the reference and wire it into your markup, and the mention stops being a loose fact floating on someone else's domain and becomes a verified extension of your own identity graph.
1. Inventory every existing mention of your name across press, podcasts, lists, and peer posts, even the ones you forgot about.
2. Check each one for consistent spelling of your name, your title, and your affiliation.
3. Flag any mention that lacks a distinguishing detail and, where you can, ask the publisher for a small edit or a linked bio.
4. Pick two or three realistic new opportunities, a podcast, a list, a peer piece, and pitch them directly.
5. Log every reference with its date, its link, and its host domain in a simple tracking sheet.
6. Re-run the audit every quarter, because old mentions age and new ones need the same disambiguation check.
A six-step pass you can run in an afternoon, then repeat quarterly.
| Reference type | Effort to earn | Corroboration strength | Typical visible lifespan |
|---|---|---|---|
| Press mention | Medium, needs a pitch and a news hook | High, if the outlet is independent of you | Long, rarely taken down |
| Podcast appearance | Medium, needs a relationship or a pitch | Medium to high, stronger with a transcript | Long, especially if transcribed |
| List or directory inclusion | Low to medium, often an application | Medium, depends on the list's own standing | Medium, lists get refreshed and names drop off |
| Peer citation | High, requires work worth citing | Highest, peers have the least incentive to inflate | Long, tied to the cited work's own lifespan |
A rough guide to where to spend limited pitching time first.
Failure modes worth naming
A few patterns quietly waste the effort people put into earning references. The first is contradiction across sources: one bio calls you a consultant, another calls you a founder, a third gets your company name slightly wrong. None of these individually look like a problem to a human reader skimming quickly, but to a system trying to reconcile facts about an entity, small mismatches accumulate into what is worth naming directly as contradiction debt, and enough of it can make a system hedge rather than commit to naming you.
The second is the orphaned mention, a reference that exists but never got linked back to you and never named you with enough specificity to be found again. These are common and usually fixable with a polite follow-up email asking for a linked credit.
The third is the citation rotation problem, where engines settle into citing the same handful of sources for a given topic and new, equally credible entrants struggle to break into that rotation regardless of merit. Understanding why that rotation forms, and how to enter it, is covered in The Citation Rotation Problem, and it is worth reading before you assume a lack of results means a lack of quality.
Why it compounds
Each reference does two things at once: it adds a trust signal today, and it ages into the record, strengthening your standing tomorrow. Network and age reinforce each other, which is why starting early matters. The references you earn this quarter are still vouching for you years from now, long after the effort behind them is spent, and every one you wire correctly into your own structured markup keeps paying that dividend without further work on your part.
There is also a knowledge graph dimension to this that is easy to miss. Public knowledge graphs like Wikidata and the systems behind Google's own knowledge panels do not build an entry for you out of thin air, they look for exactly the kind of corroborating evidence described above, independent sources that agree on who you are and what you do. A strong network signal is often the precondition for getting a durable entry in one of these graphs at all, and once you have one, it becomes another independent node that future assistants can draw on. The relationship between your own markup and these external graphs is covered in Knowledge Graphs for People, and it is worth treating as the long-run payoff of the same pitching and publishing work described in this piece, not a separate project.
FAQ
Isn't this just PR? +
Can I buy my way onto lists? +
How many references do I need? +
Does a mention need to link back to my site to count? +
What happens if two sources describe me differently? +
How does this connect to schema and sameAs markup? +
Find out what AI says about you today.
Start with a baseline. See the exact words the engines return about your name, then decide.
Claim your name →