That last part is where most estimators fall short. A number without evidence is an opinion. This page explains what a good estimator actually does under the hood, why the output should always be a range, and which inputs genuinely move the result.
What an estimator is really doing
Every domain value estimator, whatever it calls itself, answers one question: when names like this one changed hands, what did buyers pay? The differences between tools come down to three things.
- Which sales it knows about. An estimator is only as good as its record of past transactions. SoldComparables works from 128,867 recorded domain sales between May 2021 and May 2026, taken from the public NamePros "Daily Domain Sales" reports, which republish the NameBio top sales charts. Each sale carries its venue and a link to the public report it came from.
- How it decides what "like this one" means. Matching on the same letters is easy and often wrong. Matching on meaning, length, extension and style is harder and much closer to how a buyer thinks.
- How it turns a set of comparable sales into a figure. This is where ranges, confidence and venue type come in.
If an estimator cannot tell you all three, treat its output as a starting point and nothing more.
Why the answer is a range and not a single price
Domain prices are noisy. The same name can sell for very different amounts depending on who is in the room. Our corpus has several clean examples of one name selling twice.
| Domain | First recorded sale | Later recorded sale |
|---|---|---|
| zip.ai | 1,610 USD (Park.io, 2021-08-10) | 200,000 USD (Afternic, 2025-03-27) |
| agon.com | 22,100 USD (NameJet, 2023-05-19) | 249,000 USD (Atom.com, 2026-03-30) |
| reguard.com | 1,313 USD (NameJet, 2023-06-08) | 35,000 USD (Afternic, 2024-05-14) |
| mega.app | 1,350 USD (Sav.com, 2023-07-22) | 50,000 USD (Spaceship.com, 2025-06-30) |
In each case the first sale happened at an expired auction or investor venue and the second at an end-user marketplace. Nothing about the name changed. The buyer changed. A single number hides that spread. A range with a wholesale figure and a retail figure shows it.
There is also plain sample noise. umami.com sold twice at GoDaddy within five weeks: 100,000 USD (2025-12-04) and 125,700 USD (2026-01-08). Even for the same name at the same venue, two prices a few weeks apart differ by more than 25,000 USD. A good estimator reports a band around the likely value and tells you how wide it had to make it.
The inputs that actually move an estimate
Not every trait matters equally. These are the ones our estimator reads for every name, and the reason each one counts.
Extension
The extension sets the market you are selling into. In the corpus, .com accounts for 100,357 recorded sales, .org for 11,251, .ai for 4,925, .net for 3,578, .io for 1,431 and .co for 953. A .ai comparable tells you far more about a .ai name than a .com sale with the same word. Our domain comps page goes deeper on how extension narrows the field.
Meaning and term class
Is the name a dictionary word, a developer term, an AI term, an exact match keyword or a brandable? Buyers price these differently. Developer-term .ai names form their own small market: kernel.ai sold for 14,599 USD (Dynadot, 2024-10-08), workflow.ai for 35,000 USD (Park.io, 2023-01-07), webhook.ai for 2,424 USD (Namecheap, 2025-10-08).
Length, style and spelling
Shorter usually helps, but the effect flattens fast after five or six letters. Hyphens and digits change the buyer pool. Plural versus singular matters less than many people assume (rebate.com and rebates.com each sold for 500,000 USD at Sedo on 2022-12-01), but it still counts as a signal.
Venue type
Where comparable names sold tells you who was buying. In our data, wholesale venues show a median of 1,725 USD and retail venues a median of 2,288 USD, with averages of 3,312 USD and 6,437 USD. An estimator that mixes the two without labelling them gives you a number that fits neither situation.
Age of the comparable sales
Markets move. The count of recorded .ai sales went from 152 in 2021 to 2,070 in 2025. A .ai comp from 2021 is still useful, but it describes a much thinner market than one from 2025.
How confidence should work
An estimate based on twelve close comparables and an estimate based on one loose comparable should not look the same. Our rule is simple. When fewer than three good comps exist, the estimator says so, widens the range and lowers the confidence label. When no comparable recorded sale exists above 1,000 USD, it says that plainly ("names like this usually trade below that") and does not invent a figure.
That floor matters. Every sale in the corpus is at least 1,000 USD, because that is the reporting floor of the public sales reports. Most names that change hands sell for less and never show up in any public chart. An estimator that always produces a confident four or five figure number, whatever you type, is guessing.
How SoldComparables does it
When you enter a name, the tool runs these steps.
- Split and understand the name. The name is broken into words. An embedding language model has precomputed which words sit close in meaning, so "cron" is understood to be near "scheduler", "task" and "job". This lets the search find sales that share meaning, not only letters.
- Score every candidate. Each comparable gets a relevance score from 0 to 100, with up to three plain reasons, such as "Same .ai extension", "Shares the word cron" or "Same length".
- Label the venue. Each sale is tagged as "Expired auction / investor venue", "End-user marketplace" or "Mixed venue" (GoDaddy and unknown venues).
- Build the estimate. You get a low, likely and high value in USD, a wholesale figure, a retail figure and a confidence label of High, Medium or Low.
- Explain it. A plain English rationale cites the comparable sales by name, price and date, and lists what would raise or lower the number.
The result is a data-backed estimate, not a certified appraisal and not a guarantee of sale price. If you want a quick yes or no on whether a figure you have been quoted is reasonable, the domain value checker page covers that use. If you prefer to see the factors laid out as arithmetic, read the domain value calculator page.
When an estimator is the wrong tool
An estimator is built for names that resemble other names that have sold. It works well for dictionary words, common keyword combinations, developer and AI terms, and short brandables in active extensions. It is weaker for very rare one-off assets (a two-letter .com, a famous single word) where there are only a handful of comparable sales in five years. For those, the right move is to read every comparable sale closely, which is why the rationale and the source links matter more than the headline figure.
It is also not a substitute for knowing your buyer. An estimate tells you what the market has paid. Whether a specific company will pay the retail figure depends on their need, their budget and their alternatives.
Reading an estimate in practice
Say you hold a short developer-term .ai name and want a sense of value. A sensible reading looks like this.
- Look at the comparable sales first, not the figure. Are they really similar in meaning and length?
- Check the venues. If every comp sold at an investor venue, the likely value is a wholesale number.
- Check the dates. stack.ai sold for 4,449 USD (Whois.ai, 2021-07-06) and later for 72,999 USD (Dynadot, 2024-06-03). The more recent sale reflects the current market better.
- Read the confidence label and what would change the number.
- Only then decide whether to list at the retail figure, accept a wholesale offer, or hold.
Run a name through the domain appraisal tool to see the comparable sales and the range for yourself, or compare plans on the pricing page.