Domain Investing in the Age of AI: What Changed and What Didn't

01

What AI changed in demand

AI changed who buys domain names at the top of the market. The clearest signal is AI.com: Crypto.com founder Kris Marszalek bought it for $70 million, reported in February 2026, to offer consumers a personal AI agent, according to TechCrunch. The broker's announcement says the price more than doubles the previous domain record, the $30 million paid for Voice.com in 2019. AI.com now heads DNJournal's list of the largest publicly reported domain-only sales; see largest domain sales.

The same demand repriced a whole extension. The .ai namespace grew from 598,007 names on January 2, 2025 to more than one million in January 2026, according to Domain Name Wire. Escrow.com handled $10.3 million of .ai deals in Q4 2025, the first quarter above $10 million, and the average price paid for a traded .ai domain name reached about $11,000 in 2025, more than 40% above 2024, according to the Global Domain Report 2026. The largest .ai sale of 2026 so far, Bot.ai at $1,200,000, is on DNJournal's year-to-date chart. The .ai domain investing guide covers that market in detail.

AI also moved demand away from some names. Domain Name Wire's review of 2025 notes that websites that relied on Google traffic are suffering as people turn to AI tools for answers, while AI makes it much easier for an ordinary person to build a web app or technical business, which could create new demand for names. The practical reading: names valued mostly for traffic are weaker, and names a product can be built on are stronger. End users building AI products are the new buyer class.

02

What AI changed in the work

AI tools now touch every step of domain investing. Domain Name Wire's 2025 review notes that most registrars still rely on traditional keyword spinners, though many now also offer AI-powered search, and that aftermarket platforms use AI for discoverability and merchandising. It also asked when effective AI tools for outbound domain marketing would arrive, which says where the gap still is.

The tools have names in the glossary: AI domain name generators, AI-assisted outbound and portfolio pruning.

03

What AI cannot do

Four limits matter more than any feature list.

It cannot judge a buyer pool. A name's price depends on how many companies want it, how badly, and with what budget. A model sees words and past sales, not a startup's funding, roadmap or deadline.

It cannot guarantee a price. Every name is one of a kind, and one buyer at one moment sets the price. Even the new for-sale DNS standard treats listed prices as indicative only.

It cannot replace trademark checks. In WIPO Case No. D2025-2916, the panel said AI tools "may complement but are not a substitute for professional advice" and that a response which appears to be AI-generated "is unlikely to inspire confidence" in its accuracy unless the panel can see the prompts and is satisfied the text was reviewed for accuracy and truth. A UDRP complaint against a name an AI suggested is still your problem. If you might act on a borderline name, consult a qualified attorney; is domain flipping legal? explains the rules.

It cannot verify itself. AI assistants sometimes state sales that never happened. Check every comparable sale against DNJournal, NameBio or the original report before you price from it.

04

Why AI appraisals are estimates

An automated domain appraisal is a model's estimate. GoDaddy, for example, describes its appraisal as machine learning on real sales data using word tokenization, and says it is most accurate for common English-language domains. That design explains the limits of every tool in this category:

1. The training data is reported sales. Many sales are private and never reported, so models learn from a partial record.
2. Each name is one of a kind. Two names that look alike on paper can sell for very different prices.
3. Models do not see buyers. A name with two motivated end users is worth more than a model can know; a name with none is worth less than any estimate.
4. Thin data means weak estimates. Unusual brandables and young extensions have few comparable sales.

Use an AI domain appraisal as a sanity check next to comps and your own read of the buyer pool. Never quote it to a buyer as a price, and never buy a name because a tool valued it above the asking price. How to value a domain name walks through the manual method.

05

Domain flipping with AI: what is real and what is hype

Domain flipping with AI is both a real workflow and a large marketing claim. Separate them.

What is real:

• Faster screening of drop lists and marketplace inventory.
• Better first drafts of listings and outreach, including in other languages.
• Cheaper research on who might buy a name.
• Cleaner records of what you own, what you paid and what you were offered.

What is hype:

• "AI finds undervalued names for you." Everyone has the same models and the same public data. The edge is judgment about buyers, not access to a tool.
• "The appraisal shows your profit." An estimate is not an offer.
• "Bulk AI outreach sells names." Volume without targeting reads as spam and damages your sender reputation.
• "AI-generated brandables are instant inventory." If a tool can produce a name in a second, so can every other investor; see AI-generated brandables.
• "Any AI keyword plus .ai is an easy flip." Most such registrations never sell, and .ai renews two years at a time.

Domain flipping is speculative with or without AI. Most names never sell, and no tool changes that. What is domain flipping? covers how the business actually works.

06

Do domain names still matter when people ask AI assistants?

Yes. The practical view has three parts.

Traffic-based value shrinks. Direct navigation and search traffic were why many generic names earned money parked. As answers move into AI assistants, less traffic reaches websites, and Domain Name Wire's 2025 review notes that sites dependent on Google traffic are suffering. Names valued mainly for traffic lose ground.

Brand, trust and email still run through the domain. When an assistant recommends a company, the customer still lands on its domain, signs in there, receives its email and checks the address before paying. The domain is the identity layer under all of that; see machine-readable trust.

Clarity and memorability matter more. An assistant mentions a brand once, in a sentence or a voice reply. A short name that is easy to say, spell and remember passes the radio test and survives that moment; a clever spelling does not.

Be skeptical of claims that a keyword domain will make AI assistants recommend a site. Nobody outside those companies can verify how their systems choose sources. What you can work on is content and reputation, the territory of answer engine optimization and LLM citations.

07

AI agents and domains: what exists today

The standard. RFC 10023, published in July 2026 as an Informational RFC, defines the _for-sale DNS record: a TXT record under the reserved name _for-sale that signals a registered domain is available to buy. Each record must start with the version tag v=FORSALE1; and may carry one tag-value pair: fcod= (a code meaningful to cooperating parties, such as a registry and its registrars), ftxt= (free text), furi= (a web, email or phone link) or fval= (an indicative asking price with a currency code). A domain can publish several records. One example from the RFC:

_for-sale IN TXT "v=FORSALE1;furi=https://example.com/fs?d=eHl6"

The RFC is careful about limits. A listed price is "not binding" and indicative only. Publishing the record does not oblige the holder to sell. It recommends a TTL of one hour or less so stale listings expire quickly, and the signal disappears while a name is not resolving, for example during the redemption period or pending delete. It says the content is meant mainly for human-readable display, though the link and price tags are machine-parsable. And it states that automated systems should not make purchase commitments based solely on advertised prices without human verification.

Marketplace adoption. In September 2026, the marketplace Atom made two announcements. On September 16, it said AI agents can search, price and buy domains on Atom in one programmatic flow through the Machine Payments Protocol (MPP), an open standard from Stripe and Tempo, covering new registrations through its registrar and premium names on its marketplace. On September 25, it said it is rolling out RFC 10023 records for listings that use its nameservers, with a link to the listing, the price where a fixed Buy It Now price exists, and a short note; make-offer and price-on-request listings publish no price.

That is what exists today: a published convention, a marketplace publishing it, and agent checkout on that marketplace.

What could follow (speculation, not fact)

None of this is established. Treat it as scenarios to watch.

• Price discovery by machines. If more sellers publish records, software could compare asking prices across thousands of names in minutes. Names with a clear asking price would be easier for an agent to act on than make-offer names.
• Public prices become anchors. A price in DNS is visible to every competitor and scraper, so some sellers may prefer to publish a link and no price.
• Spam and fake signals. The RFC itself warns that the record can be abused as a marketing lure. Buyers and their agents will still need WHOIS or RDAP checks and escrow.
• Agents as a buyer class. Whether AI agents become a real source of demand, or stay a checkout convenience for people, is unknown; see agentic domains.

08

The 2026 new gTLD round

ICANN's application window for new generic TLDs ran from April 30 to August 12, 2026, the first round since 2012, which introduced more than 1,200 new gTLDs. ICANN received more than 1,600 applications, and 1,616 will proceed. Reveal Day, when ICANN publishes the applicants and the strings applied for, is October 7, 2026; String Confirmation Day is November 17, 2026; objections and community input run from November 17, 2026 to March 16, 2027; a prioritization draw follows in the first half of 2027. ICANN has not announced launch dates; see new gTLD round (2026).

For investors, a new round adds choice, not scarcity. New gTLDs held 52.9 million registrations at the end of June 2026, up 34% in a year, against 166.6 million for .com alone, according to DNIB. Watch Reveal Day for AI-related strings, but do not build a strategy around extensions that do not exist yet.

09

What did not change

• Scarcity. Every domain name is one of a kind. AI can generate a million candidates; it cannot create a second Bot.ai.
• Buyers. A name sells when a person or company with a need and a budget wants it. An agent that buys a name acts for someone.
• Patience. Good names can wait years for the right buyer, and no tool shortens that wait.
• Renewals. Every name costs a renewal fee every year, every two years for .ai, whether or not it sells. AI lowers research costs, not holding costs.
• Trademark law. The UDRP's three elements are unchanged, and WIPO handled a record 6,282 domain disputes in 2025, according to Domain Name Wire.
• Transaction basics. Escrow, auth codes and transfer locks work as before; a newly registered name typically cannot move to another registrar for 60 days.

Selling still works the way how to sell a domain name describes. AI makes the paperwork faster; it does not make the buyer appear.

10