What AI Citability Means for an Expired Domain

The question people used to ask about an expired domain was "what's the DA". The question I get now is some version of "will ChatGPT trust it".

Nobody can answer that one directly, and anybody who says they can is selling something. But the question underneath it is real, and it has a better shape: does this domain sit in the part of the web that answer engines draw from? That is a question about provenance, and provenance is the one thing an aged domain either has or does not have. You cannot manufacture fifteen years of being referenced by other people.

Why provenance rather than rankings

Answer engines do not work like a results page. They assemble an answer from sources, and which sources they reach for is not a straightforward function of who ranks first.

Two measurements make the point. In a study of 22,295 AI answers to business-software prompts, the ten most-cited domains accounted for only 11 to 13% of citations on each engine, leaving around 88% of citations spread across everything else. A separate measurement study from Washington University, covering 55,393 trending queries, found that nearly 30% of the domains cited in a Google AI Overview do not appear anywhere in the first-page results shown beside it.

Read those together and the picture is awkward for anyone trying to optimise their way in. Citations are dispersed, they vary by engine, and a meaningful share of them go to sources that were not winning the ranking contest at all. There is no top-ten position to buy your way into, because the top ten is not where most of the citations live.

Our working assumption is that the sources engines reach for tend to be ones other credible sources already reference, and that a domain's inbound provenance is therefore worth measuring. That is a premise rather than a finding. No published study we have found tests it directly, which is a gap we come back to below. What it is not is a ranking signal you can acquire this quarter.

What the score is

Agent Citability is a 0 to 100 score for how deeply a domain sits in that trusted layer of the web. It is computed by seeded trust propagation: trust flows outward through the link graph from a curated, tiered list of over a thousand sources, following the TrustRank method published by Gyöngyi, Garcia-Molina and Pedersen in 2004. The current release covers more than 100 million domains, and final scores are calibrated to a percentile within each release.

The tiering is the part that does the work. A reference from an encyclopedic, governmental or academic source counts for a great deal, because those references are rare and hard to obtain. A reference from a newsroom or a reference publisher with editorial control over its outbound links counts for less but still counts. A link from a platform where anyone can place one in about a minute counts for close to nothing, and that includes some of the platforms answer engines cite most often. That looks like an oversight until you think about what a score would be worth if a link you could create yourself moved it.

The methodology is published: the tier philosophy, example seed domains and the release process. The weights are not, and we would rather say that plainly than imply an openness we do not have. You can audit what the metric claims to measure and argue with how the tiers are drawn. You cannot recompute the number. The methodology page sets out both halves and is blunt about which parts are validated.

Reading it on an expired domain

Here is where it gets specific to aged domains, and where I think the metric earns its place next to the older numbers.

A live business can work on its citability. It can publish research, get quoted, earn a reference from a standards body. Slowly, but it can move. An expired domain has stopped doing any of that. Nothing new is being earned, so what the score reads is what other people decided about the name while it was alive.

The score itself is not frozen. It is recomputed with every release like any other, and on a dead domain the movement is mostly downward as the pages holding those old references get deleted or redesigned. What is fixed is the underlying behaviour: no campaign is running, no PR team is working the number, and nobody is adding references to a site that no longer exists.

So a high score on an expired domain is telling you something quite narrow and quite valuable: at some point, sources the tiering rates highly chose to reference this name. That is the property you are acquiring. Everything else about the domain, the traffic it used to get, the rankings it used to hold, is gone.

A few practical readings:

A high score with a short referring-domain list is the profile I look at hardest. It suggests a handful of strong references rather than volume, which is the shape that count-based metrics undersell. Whether that makes it underpriced depends on everything else you check.

A high referring-domain count with a low score is the profile to look at twice. It is consistent with volume from places the tiering does not rate, which is the 412-referring-domains trap from evaluating link history showing up as a number. It is also consistent with links whose host pages have gone quiet, since recency and reachability both feed in. Either way the answer is to open the linking domains rather than to trust the gap between the two figures.

No score at all is not a zero. It means we hold no score for that domain in the current release, which is a different claim from having measured it and found nothing. Coverage is uneven, and plenty of perfectly good regional domains have no score for that reason alone.

What it does not tell you

The score does not predict that any engine will cite the domain. It measures a capacity, not an event, and the citation record is dispersed enough that a per-domain prediction would be fiction.

It says nothing about content accuracy, yours or the previous owner's, and it says nothing about outcomes. Provenance is an input. Traffic, rankings and citations are results, and the distance between the two is where all the actual work happens.

There is also a gap worth naming. Published research measures which domains get cited. We have not found a study that tests whether a link from a trusted domain predicts that the linked domain gets cited, which is the hypothesis the whole metric rests on. Until that test runs properly, the tiering is a well-grounded prior rather than a validated model, and citation behaviour is noisy enough that anyone claiming a tight correlation deserves a raised eyebrow.

Scores fall

This is the part that surprises people who are used to authority metrics, which only ever go up.

Scores are recomputed for each rebuild of the graph and shipped as versioned releases. A domain drops when trusted sources stop referencing it, when a linking domain is re-tiered, or when recalibration shifts the percentile underneath it. A falling score is usually the metric working correctly: trust was withdrawn somewhere and the number noticed. Scores are comparable within a release, and the release notes exist so you can see what moved.

For an expired domain, the practical implication is that the links behind the score are also decaying. Pages get deleted. Sites get redesigned. A domain that has been dead for eight years is losing references quietly the whole time, which is an argument for looking at the archived history alongside the score rather than treating the number as permanent.

Where to look

The free citability checker will look up any domain we hold a score for, listed here or not. Every listing in the Revised directory shows its score alongside the trusted sources behind it, so the number arrives with its evidence rather than on its own. Sorting and filtering the whole directory by citability is a paid feature; what each plan includes is set out there.

If you want the background on how answer engines choose sources in the first place, content cited by authoritative sources and the guide to generative engine optimisation cover the wider picture.