Cold email domain generator and fleet planner

Results

AI-reviewed names must pass separate relevance and natural-phrasing checks before selection. The displayed quality score compares names within that set; it does not predict inbox placement. The standalone scorer below uses the original naming heuristic.

Score a domain you already own

Paste a sending domain you already use, or one you are about to buy elsewhere, and see it against the same census. This is where the full range of the score shows up.

How many sending domains do you need?

Domain count is a consequence of volume, per-mailbox sending rate and how many mailboxes you put on a domain. The rates below are what ReplyLead actually runs, not provider maximums. For the volume question itself see how many cold emails per day.

Why a separate domain, and why not variations of it

Our rule at ReplyLead: the primary company domain never sends cold email, on any campaign we run. Cold outreach carries bounce risk, complaint risk and listing risk, and if the domain your business runs on absorbs that, you lose the mail the business depends on: invoices, support threads, password resets, replies from customers.

Almost everyone gets that far. The part that gets skipped is what the secondary domains are called. The standard advice, and the output of the cold-email domain generators we reviewed, is variations on one stem: getbrand.com, brandmail.com, brandhq.com, try-brand.com. Tidy, easy to manage, and instantly recognisable as a pattern, which is exactly the problem: a pattern legible to you is legible to a filter.

The rule we run on now: if you can describe your domain set in one sentence, so can a filter. Vary the stem, not just the suffix, and vary it across provider pools too, so that a Google-pool pattern and a Microsoft-pool pattern do not share a signature. That is what the generator above is built to do, and what the fleet analysis measures. The wider architecture, including provider mix and pool isolation, is covered in cold email infrastructure.

What 12.9 million business domains actually look like

The suitability score on this page is not a taste judgement. It is calibrated against a census of 12,915,691 business domains observed in ReplyLead contact data, each one resolved in DNS. Published here because we could not find this measured anywhere else, and because a generator that tells you a name is "too long" should be able to say what it is long compared to.

Shape of a real business domain

Naming structure and domains that stopped resolving

Every domain in the census was collected because it belonged to a real business at the time it entered our contact data. Some no longer resolve in DNS at all, which means the registration lapsed or was abandoned. Cross-tabulating that against name structure produces the finding below, on a reliably-resolved subset of 12,335,726 domains where the baseline no-longer-resolves rate is 9.26%.

The result worth keeping. One hyphen is close to neutral, at 1.04x the baseline. Two or more is not, at 1.75x. The same shape holds for length: the association is monotonic across every band, from 6.42% at six characters or fewer up to 14.26% at twenty-three or more. A digit sits at 1.26x. So the common advice to avoid hyphens entirely is, on this population, slightly overstated for a single hyphen and clearly understated for two.

Limitations, stated plainly

• This is observational and confounded. Domains entered the dataset at different times, so a longer observation window is one plausible driver of a higher lapse rate. Nothing here establishes that a naming choice causes abandonment.
• "No longer resolves" means NXDOMAIN in DNS. It is a proxy for a lapsed or abandoned registration. It is not a measure of spam, of blocklisting, or of deliverability.
• The population is business domains observed in ReplyLead contact data. It is not a random sample of the global DNS, and it is weighted toward the markets we sell into.
• One further census bucket, covering domains recorded as publishing no usable records, was excluded from the cross-tabulation: when we re-resolved a sample of it live, only 25.8% still matched. Excluding it is why the cross-tab denominator is 12,335,726 rather than the full 12,915,691.
• Structural measures such as length, hyphens and digits are properties of the name itself and do not go stale. Only the resolution state does.

How current this is

On 30 August 2026 we re-resolved a fresh sample of 1,600 domains against live DNS to measure how well the stored census still holds. Domains recorded as no longer resolving were confirmed at 98.0% (392 of 400). Domains recorded as having a working mail record were confirmed at 99.2% (397 of 400). The bucket that failed re-verification is the one excluded above.

Methodology

What availability means here

Registration status comes from registry RDAP and WHOIS responses. The tool reads the IANA bootstrap registry to find the authoritative RDAP server for each ending, then queries it directly from your browser. A 404 means that RDAP service returned no domain object, which is why the tool reports appears unregistered rather than "available". A 200 means it is registered. A rate limit, a timeout, a server error or a missing RDAP service is reported as could not verify, and never as unregistered. For .co, .us and .me, ReplyLead queries the official WHOIS registry server. If a browser RDAP request fails, ReplyLead also attempts the registry lookup server-side. With registration checking enabled, only names with an explicit no-registration-record response enter the result list. Inconclusive checks are excluded; repeated failures pause the search with a retry option. Checks run at bounded concurrency with pacing between requests, and conclusive results are cached for up to five minutes to avoid repeatedly querying the same name.

No registration record is not the same as purchasable. A registry can hold a name back as reserved, restricted, blocked or premium-priced without any registration record existing for it, and some names carry eligibility rules or a price far above standard. RDAP answers one narrow question, which is whether a registration object exists right now. Confirm every name at a registrar before you buy it. Registration is also continuous, so a reading taken now can be stale minutes later. The IANA bootstrap currently lists RDAP servers for about 1,200 endings. For endings such as .co, .us and .me without an IANA-listed RDAP endpoint, this tool uses the official WHOIS registry through ReplyLead. An unsupported or failed lookup is never counted as proof of non-registration.

How the tool works out what your business does

If you enter a website, we read it server-side and take only the high-signal parts: the page title, meta description, H1 and H2 headings, navigation labels, structured data and the opening hero copy, across the homepage and up to two pages whose links say services, products, capabilities or about. Whole pages are deliberately not ingested, because boilerplate is exactly what makes every company look like "business solutions". The website is read by a ReplyLead endpoint. With AI review enabled, the extracted business context is also used by Google Cloud and Typesafe for name proposals and evaluation. Explicit service descriptions you enter take precedence when they conflict with the website.

AI review combines generated name ideas with your business terms, semantic neighbors and observed word pairings. It automatically requests fresh proposals using vocabulary not yet represented in your list. It evaluates relevance and natural phrasing separately, then selects names under the strict no-repeat rules. Candidate ideas rejected by quality or registration checks are replaced from the remaining pool. The quality thresholds stay fixed when more names are requested. The initial pool is expanded as the search continues. Exhausting a pool does not complete the request: the tool keeps working toward your requested count. If repeated expansion produces no further valid names, it pauses with a Continue search option and shows how many are still needed.

Confidence controls generation. Evidence is counted: a website we could actually read, a description you wrote, an industry, keywords, and how many specific concepts came out. With high confidence the tool generates normally. With medium it generates at most 50 names and caps them at 82. With low confidence it does not generate a normal set at all: it says so and asks for a description, an industry or a few keywords, because names invented from generic vocabulary would look plausible and mean nothing. A name can never score higher than the evidence behind it.

What outbound suitability scores, and what it does not

Outbound suitability is a naming and infrastructure heuristic. It does not predict inbox placement. No property of a domain name determines whether mail is delivered; authentication, sender reputation, list quality, volume and content do that. The weights are: semantic relevance 30, business fit 20, professionalism 15, readability 10, fleet diversity 10, spelling risk 5, length 5, pattern risk 5. Length is deliberately minor, because a name that says what you do is usually longer than one that says nothing. The length curve, the hyphen rule and the digit rule are all set from the census above rather than chosen by feel: 92.7% of business domains carry no hyphen, 97.0% carry no digit, and the median name is 11 characters.

Relevance is a gate, not just a weight. Every candidate is built from concepts read out of your company name, description, industry and keywords, and it carries a record of which concept each half came from. A name scoring under 50 on relevance is capped at 69 overall no matter how well it reads, and one scoring under 30 is never shown. That is why you will not see a short, pretty, meaningless name near the top: short and readable cannot outvote irrelevant. The same precedence applies to promotional wording, names with no vowels, two or more hyphens, multiple digits, and names longer than 99.5% of the census.

How results are ordered

Names are selected for relevance and readability subject to the mandatory brand exclusion, word-family uniqueness and name-containment rules. A requested count does not relax those constraints.

Brand-containing names are excluded. Each normalized word family can appear in only one returned name, regardless of position; business prefixes and suffixes count too. A name cannot contain another returned name. These constraints are mandatory. If too few relevant candidates pass, the tool reports the shortfall without adding branded or repeated-word alternatives.

The search continues through additional industry vocabulary and naming combinations until it finds your requested count or exhausts its relevant vocabulary. Found domains remain in the list while it searches for the rest. You can pause and continue. Only a registry response with no registration record counts as found; unverified lookups are counted separately.

How fleet similarity is calculated

For the set you select, the tool computes pairwise character-bigram similarity across every pair, the longest shared prefix and suffix, the number of domains sharing a common word token, and the share of the set sitting on a single ending. The headline diversity score combines those into one number where high is good. It is transparent lexical analysis on the names alone. It has no access to registration dates, registrars, hosting or reputation data, and it makes no claim that a given score produces a given delivery outcome.

Where the capacity numbers come from

The planner uses ReplyLead's current operating rates, measured across our own live mailboxes: roughly 10 to 12 cold sends per Google Workspace mailbox per day and roughly 1 to 2 per Microsoft 365 (Exchange Online) mailbox per day, with 1 to 3 mailboxes per domain in the Google pool. These are operating choices, not provider ceilings. Google's own documented limits for paid editions are 2,000 messages and 3,000 external recipients per user per day, both far above anything appropriate for cold outreach, and a technical ceiling is never a volume recommendation. Plan on two to three weeks of warmup per mailbox before real campaign volume. Warmup itself is covered on email warmup.

What the generator will not produce

No filler. AI proposals may introduce additional service vocabulary, but every AI-reviewed name must pass separate relevance and natural-phrasing checks before it can enter your list. No homoglyph or lookalike substitutions, no deliberate misspellings of your brand or anyone else's, and no names constructed to make a recipient believe the mail came from a different company. The company stem is excluded from generated names. Impersonation is not, and no setting turns it on. The tool does not screen for trademarks; that is a check to run with counsel before you register anything.

Privacy

There is no signup. With AI review enabled, the business details and candidate names are sent through ReplyLead to Google Cloud for name proposals and Typesafe for relevance and natural-phrasing evaluation. ReplyLead temporarily caches proposals and scores for up to one hour to avoid repeat requests. Standard review uses the original naming checks. Website reads and registration checks use the services described below.

In short: website reads use ReplyLead; AI review sends business context and candidate names to Google Cloud and Typesafe; registration checks use domain registries directly or through ReplyLead. API credentials are kept on the server.

Questions people actually ask

Next steps once the domains exist

Generate SPF and DMARC records for each new domain  //  Cold email infrastructure, the pool architecture these domains sit in  //  Email warmup before real volume  //  How many cold emails per day  //  Mail provider lookup  //  Cold email deliverability  //  Deliverability checker  //  All complimentary tools

Built and methodology reviewed by Mark Glazer, founder of ReplyLead. Census and tool last updated 30 August 2026. Our standards for what we publish and how to request a correction are on editorial standards.