This page explains each step: where the sales come from, how they are labelled, how similarity is measured, how the estimate and its confidence are built, and where the method stops. If you rely on our numbers in your own work, read this first.
Where the sales data comes from
The corpus holds 128,867 recorded domain sales, from 28 May 2021 to 28 May 2026. Every sale comes from the public NamePros "Daily Domain Sales" reports, which republish the NameBio top sales charts. For each sale we keep the domain, the price in USD, the date, the venue and a link to the public report thread it appeared in.
That link is shown next to every comparable in the product, so you can open the original report and check the sale yourself. We do not invent sales to fill gaps. The latest recorded sale in the data is dated 28 May 2026. We describe the data as "latest recorded sales" with their dates, not as a live feed.
The 1,000 USD floor
The public reports list sales of 1,000 USD and above, so every recorded sale in the corpus is at least 1,000 USD. This matters. The median recorded sale is 1,900 USD and 86.8 percent of recorded sales are below 5,000 USD, but the much larger market of hand registrations and small resales below 1,000 USD is not in the data. When a name has no comparable sale above the floor, the honest answer is that names like it usually trade below 1,000 USD, and that is what we say.
Dates
Every comparable shows its sale date. Prices move with the market, especially in fast-growing extensions. Recorded .ai sales went from 152 in 2021 to 2,070 in 2025, for example, so a .ai sale from 2021 tells you less about current buyers than one from 2025. You can filter comps by date range, and the estimate gives more weight to recent sales.
Venue tiers
Where a name sold says a lot about who bought it. We label every sale with one of three tiers, using plain labels.
| Tier | Label in the product | Venues | Recorded sales | Median |
|---|---|---|---|---|
| Wholesale | Expired auction / investor venue | DropCatch, NameJet, Park.io, Catched.com, Dynadot, Namecheap, Sav.com, Whois.ai, NamePull.com | 30,021 | 1,725 USD |
| Retail | End-user marketplace | BuyDomains, Afternic, Sedo, Atom.com, DAN.com, DomainMarket.com, Spaceship.com, private sales, brand marketplaces | 23,816 | 2,288 USD |
| Mixed | Mixed venue | GoDaddy and venues we cannot classify | 75,030 | 1,800 USD |
The gap is real. The average retail sale in the corpus is 6,437 USD against 3,312 USD wholesale. The same name can sell at both levels: zip.ai sold for 1,610 USD on Park.io (10 August 2021) and then for 200,000 USD on Afternic (27 March 2025), and agon.com sold for 22,100 USD on NameJet (19 May 2023) and then for 249,000 USD on Atom.com (30 March 2026). An estimate that mixes the two without saying so is misleading, so we keep them apart. More on this in wholesale vs retail domain prices.
How a name is analysed
Before searching, we normalise what you type (lower case, no scheme, www or path, and .com when no extension is given) and describe the name with a set of features:
- length of the name before the extension;
- the TLD, such as .com, .ai or .io;
- singular or plural, with the singular form;
- term class: developer term, AI term, dictionary word, brandable or exact match keyword;
- style: exact match, brandable, two-word, acronym or numeric;
- whether it contains digits or hyphens.
For crons.ai, that gives a 5 letter .ai name, plural, singular form "cron", developer term.
How similarity is scored
Words, not just letters
Every sold name in the corpus is split into words, so "workflowhub" becomes "workflow" and "hub". An embedding language model has precomputed which words are close in meaning. For example, "cron" sits near "scheduler", "task" and "job". This lets the search find sales that share meaning with your name, not only sales that share letters. A keyword search for "cron" finds almost nothing, while a meaning search also surfaces developer-term .ai sales such as stack.ai, server.ai and kernel.ai.
Relevance and reasons
Each candidate sale gets a relevance score from 0 to 100, built from closeness in meaning, the same or a related extension, similar length, the same term class and style, singular or plural form, and recency. Next to the score you see up to 3 plain reasons, such as "Same .ai extension", "Shares the word cron" or "Same length". If you cannot see why a sale was chosen, it should not be in your report, so the reasons are always shown.
You can narrow the result with filters for TLD, price range, date range, venue tier, style and maximum length.
How the estimate is built
The range
The estimate uses the most relevant comps, weighted by relevance and recency. From them we derive a low, a likely and a high value in USD. The low and high values are not the cheapest and dearest comps. They describe where most of the weighted comparable sales fall, so a single outlier sale does not stretch the range.
Wholesale and retail figures
We compute a wholesale figure from the comps sold on investor venues and a retail figure from the comps sold on end-user marketplaces. If you are bidding at an expired auction, the wholesale figure is the one to watch. If you are pricing a name for an end user, look at the retail figure.
Rationale and movers
Every estimate comes with a short rationale in plain English that cites the comps by name, price and date, for example "server.ai sold for 29,400 USD on Namecheap on 25 February 2025". It also lists what would raise the value (a buyer in the exact niche, a shorter or singular form) and what would lower it (plural form, weak extension, few recent sales).
Confidence
Confidence is High, Medium or Low. It depends on how many good comps exist, how close they are, how recent they are and how much their prices agree. Several close, recent comps with similar prices give High. A few loose or old comps, or comps with prices far apart, give Low.
When comps are thin
If there are fewer than 3 good comps, the result says so, widens the range and lowers confidence. If there is no comparable recorded sale above 1,000 USD at all, the result says that names like this usually trade below that and gives no figure. We do not fall back to a formula or a guess.
What we do not do
- We do not invent sales or fill gaps with synthetic data.
- We do not claim a live feed. The data runs to the latest recorded sale, dated 28 May 2026.
- We do not show a number without the sales behind it.
- We do not use your searches to buy names. Searches are private to your workspace.
Known limits
Reported sales are a sample of the market, not all of it. Many sales are private and never reported, and some venues report only part of their sales. Nothing below 1,000 USD is included. Prices are in USD as reported, without fees or currency adjustments. The corpus is dominated by .com (100,357 of 128,867 recorded sales), so extensions such as .io (1,431) or .co (953) have fewer comps and lower confidence. An estimate is a data-backed opinion, not a certified appraisal and not a guarantee of sale price.
Corrections
If a recorded sale looks wrong, for example a wrong price, date or venue, write to [email protected] with the domain and the source link. We check it against the public report and correct or remove it.
How SoldComparables puts it together
You enter a name, the engine describes it, finds candidates by meaning and features, scores and explains each comp, then builds a range, a wholesale and retail split, a confidence level and a rationale. You can see the same steps in our guides to domain comps and comparable sales.
Run a search on the domain appraisal tool to see the method on your own name, or compare plans on pricing.