How we verify.

Sourced facts

Every fact comes from a first-party source: the vendor’s own pricing page, documentation, download page, changelog, or announcement. We fetch the page and read what it says — its text or the structured data it publishes. Nothing is written from memory, third-party coverage, or a model’s guess.

Facts follow a fixed vocabulary — delivery, platforms, editor integrations, business model, free tier, single sign-on, enterprise offering, documentation, and changelog — so tools can be compared like for like.

Receipts

Each fact, price, and change carries a receipt: the page it came from and the date we checked it. Open one to see the method, confidence, what the page said, and the link.

Observation dates

A fact is true as of the date it was observed, not forever. Prices and plans change, so they are recorded as dated observations in an append-only history: a new observation is added; nothing is overwritten.

Freshness

Every record shows how recently it was checked:

• Fresh — verified within 30 days.
• Aging — 30 to 90 days.
• Stale — more than 90 days. Treat with care.

Unknown stays unknown

When we have not verified something, we say so: Not yet verified. A missing value is never shown as blank, as zero, or as a plausible guess. Custom “contact sales” pricing has no number, so it is described in words, not priced.

Change logging

A change is a dated event documented by a first-party page — a price or plan change, a release, a platform change, availability, a rebrand, or an ownership change. Re-checking a value and finding it unchanged is not a change: it refreshes freshness and never enters the change log. Price history is charted only when a price actually moved; until then a profile says “History starts here.”

Market pulse and counters

Activity counters — changes in the last 7, 30, and 90 days, tools changed, pricing changes, “most changed” — are counts of logged changes over whole UTC calendar days ending today. Changes are dated to the day the vendor’s page gives, so there is no hour-level window. A count of logged changes reflects the pages ToolMarketCap reads, not a vendor’s total activity: more changes is not better, and “most changed” is not a ranking or a trend.

Changes are grouped into four kinds for these surfaces: pricing & plans, releases, availability & platforms, and ownership & names. Colors mark the kind, never whether a change is good or bad.

Market readings

The readings on the overview, segments, profiles, pricing, and comparisons are fixed statements over the same verified record: counts of logged changes, shares of a stated whole, and medians over a stated set. Each one links to the records it counts — usually the change feed, filtered to exactly those changes.

• Windows are the last 7, 30, or 90 whole UTC days, today included. “The previous 30 days” is the equal window just before. It is compared only when ToolMarketCap’s direct tracking covers all of it for every tool counted — each profile’s “tracked since” date, its first direct observation — and only as two counts side by side, never a percentage change. Changes read back from a vendor’s changelog show that a record exists for a period, not that ToolMarketCap was observing it, so they never establish coverage on their own.
• Shares (“11 of 33, 33%”) show a percentage only when the whole holds 20 or more changes; below that, counts only. A segment’s share is shown beside its share of tracked tools, since segments differ in size.
• Entry-price medians use the verified monthly entry price (lowest verified paid plan billed monthly, USD) and appear only when at least 3 tools covering 50% of the set have a verified monthly entry price — always with the count they are taken over. Tools without one are explained (free plan only, no paid plan billed monthly, no verified price), never counted as $0. Annual, usage-based, and custom prices are never converted.
• Against the segment: a tool’s 30-day change count is set beside its segment’s per-tool median (segments of 3+ tools), and its Activity beside the segment’s Activity median. A limited-history Activity value is a lower bound, so it is only ever said to be above a median.

Readings describe ToolMarketCap’s record. They do not measure adoption, popularity, usage, revenue, or quality, and “more changes” or “above the median” is not better. A quiet window can mean the vendors changed little, that their pages published little ToolMarketCap could read, or that recent changes are still being verified; the latest verified change is dated wherever recent windows are empty.

Weekly briefs

A weekly brief covers one ISO week, Monday to Sunday (UTC). It is assembled deterministically from the changes logged that week and the dates tools were first observed. Every sentence is a fixed template filled with counts and names — for example “Coding agents logged 5 verified changes, including 4 releases.” Nothing is generated, estimated, or interpreted. A week with fewer than three logged changes still has a page but is not listed or indexed; the current week is marked in progress.

Price history and archived captures

Every price observation is kept. A price movement is two consecutive observations of the same plan at different list prices; it is dated by the first observation at the new price, and both observations and their receipts are shown, because the change happened somewhere between them. Direction is stated as an increase or a decrease, never as good or bad.

Earlier prices may be read from archived captures of the vendor’s own pricing page in the Internet Archive. The receipt links the exact capture, and the observation is marked “Archived capture”. Archived captures extend price history only; a tool’s “tracked since” date is the first day ToolMarketCap observed it directly.

No quality score, no ranking

A quality score is a claim about which tool is better, and a ranking is a claim about order of merit. ToolMarketCap has no data that could defend either, so it publishes neither: no quality score, no “best” list, no trending list, no momentum score, and no adoption, traffic, or revenue estimate.

The one index it does publish, the Activity Index, measures something it can observe: how much verified, dated change activity a tool has shown recently. It is versioned, its formula is public, every value is explained from the records behind it, and a higher value is not better. Sort orders on the market board are factual — most recently changed, price, name.

Markets

ToolMarketCap classifies professional AI tools into a small set of markets — AI Coding, Assistants & Research, Work & Productivity, Automation & Agents, Creative AI — and each market into segments. Every tool sits in exactly one segment of one market. Markets are an editorial classification, listed in a fixed order and never ranked; a market appears only once it holds a released tool. Foundation models (GPT, Claude, or Gemini model families) are a different asset class and are not tracked as tools; the products built on them are.

A product is tracked when most of these hold: it is a professional, active commercial product from an identifiable vendor; its own first-party pages document it, its pricing or commercial model, and its changes; it is durable and comparable with other tools in a real market; and the evidence supports a trustworthy canonical record. Novelty demos, thin wrappers, abandoned products, and products without first-party evidence are not tracked — a candidate whose first-party pages cannot be read is left out rather than filled in from elsewhere.

Every tool carries the same core facts — delivery, platforms, business model, free tier, single sign-on, enterprise offering — and its market adds the few that only mean something there (editor integrations in AI Coding, workspace integrations in Work & Productivity, app integrations in Automation & Agents, outputs and commercial use in Creative AI, and API access in every market except AI Coding, whose record has not researched it). Free tier has one meaning in every market: a free plan that does not expire. A free trial, or credits that run out after a set period, is recorded as no free tier; where the vendor’s own pages do not settle it, the fact stays “Not yet verified”. Two tools are compared on the core, plus their market’s facts when they share a market. A market new to the record shows its tracking start date and limited history until direct tracking covers the periods it is compared over.

Segments and peers

Segments are an editorial classification inside a market: in AI Coding by how each tool is delivered (editor, terminal agent, extension, browser builder, cloud agent); elsewhere by the job it does (meetings, workflow automation, video). Every tool in a segment is a peer of every other, and comparisons are only generated within a segment. Segments are labeled as editorial wherever they appear and assert nothing about quality or size.

Comparisons

Any two tracked tools can be compared, and a comparison counts only verified facts: differences first, then what both share, then what is not yet verified for one or both. Links such as documentation are shown but never counted. A comparison is listed and indexed only when the two tools are peers, each has at least four verified comparable facts, and the record shows at least two real differences. Other pairs still open, but are not promoted.

Alternatives

A tool’s alternatives are the tools in its editorial segment (its use case) plus its editorial peers — classifications already shown on every profile. Nothing else makes a tool an alternative, and no text similarity or model is involved. They are listed alphabetically and never ranked.

Every alternative carries why it appears: each condition read as met, not met, or not verified, with the first-party page behind it — same use case, a verified free tier, an open-source licence, self-hosting, SSO with an enterprise plan, a different delivery surface, recent verified changes, or a lower verified entry price. Entry prices are compared only when both are monthly list prices in USD for the same unit (whole plan, seat, or user); anything else is “not comparable”, never converted. Pages such as /alternatives/zapier/self-hosted exist only when at least one alternative verifiably meets the condition, and list separately the alternatives the record says nothing about.

A base alternatives page is indexed only when the tool has at least four verified comparable facts and at least three of its alternatives do too; a filtered page needs an indexed base page and two documented alternatives that meet its condition. Thinner pages are reachable but not indexed.

Switch from a tool

Switch checks a tool’s alternatives against the reasons you choose — too expensive, a free option, open source, self-hosting, enterprise controls, a different deployment model, a current offer, or a more actively changing product. Each reason is one stated condition from the list above. Results are grouped as meets every requirement, not ruled out — some not verified, and does not meet, ordered by how many of your requirements are met, then by name. The same selection always gives the same result. Nothing is scored, and no answer is generated by a model.

Screener

The Screener filters the record by verified values only. A tool whose value is not verified never passes a filter that needs it; it is counted as “not verified”, not hidden. Prices are compared only on one price basis — monthly, in USD, per plan, per seat, or per user, chosen on the screen: the price filter applies only to prices on that basis, tools priced on another basis are counted as “priced on another basis”, and the price sort lists each basis as its own group, never interleaved. No price is converted and no seat count is assumed. Sorts each have one meaning: name; verified monthly entry price (tools without one last); verified changes in the last 30 days; when the canonical record last changed; and when ToolMarketCap first verified the tool. There is no popularity sort, because ToolMarketCap has no popularity dataset. A screen’s whole state is in its URL.

Open source and self-hosting

Open source means an OSI-approved licence read from the vendor’s own repository or page. Code that is public under another licence (such as n8n’s Sustainable Use License) is source-available, which does not meet an open-source requirement. Self-hosting is recorded only when the vendor states the product runs on your own infrastructure; a CLI that runs on a laptop is not counted as self-hosting unless the vendor says so. A “No” is recorded only when the vendor says so. Everything else is Unknown. Repository stars are not collected or used.

Market Activity

Market Activity (tmc-market-activity-v1) shows where observable product activity is happening, market by market: verified changes in the window, of which pricing and plan changes, how many tools changed (breadth), and how many tools began tracking. Markets are ordered by verified changes, then breadth — a count, not a ranking. A market’s week is above usual when it holds at least five changes and at least twice the market’s median week over the previous eight weeks; a market with under four weeks of record is not assessed.

It is a separate methodology from the Activity Index and does not change it. It reads no page views, social signals, vendor payments, affiliate data, or offers, and it says nothing about product quality, company value, or investment merit.

Market Heat

Market Heat (tmc-market-heat-v1) puts one of five states on each market, read from Market Activity and never changing it. Above usual is Market Activity’s own pace rule (this ISO week holds at least five verified changes and at least twice the market’s median week over the previous eight). Active: this week already holds more verified changes than the median week, and at least two. Normal: changes in the last 30 days, this week at or below the median. Quiet: no verified change dated in 30 days. Not assessed: under four weeks of record. Every state is shown with the numbers behind it. Heat is not investment heat, valuation momentum, or demand.

Market Signals

Market Signals (tmc-signals-v1) are separate families, never combined into a score: markets above their usual week; pricing moves (observed list-price increases and decreases, plus verified pricing and plan changes); unusual change activity — a tool with at least three verified changes in 30 days and at least twice its own median over the previous three 30-day periods, tracked for at least 120 days, so each tool is measured only against its own record; availability, platform, ownership, and name changes; changes across at least three change areas; open-source and self-hosted products with a change; newly tracked products; and, apart from all of these, new verified offers. Each family states its window, calculation, order, and why an item appears, and every item links to its record. More change is not better, and no signal is a recommendation.

Change momentum. ToolMarketCap does not add a second momentum score: a tool’s observable product activity is the Activity Index, and “unusual change activity” above is that tool against its own past.

Weekly Market Pulse

Each ISO week (Monday to Sunday, UTC) has a permanent page at /pulse/week/yyyy-wNN: verified changes, observed list-price increases and decreases, verified pricing and plan changes, newly tracked tools, and each market’s heat as of the last moment of the week. Every figure is a count of dated records, so a past week does not depend on anything’s current state; only a correction, or a dated record added later, can change it. Offers are not part of a week: the record holds their current status, not a dated history, and a past count would drift when one ended or was withdrawn. Current offers are on the offers page. A week is indexed once complete and with at least three changes.

Comparisons as shareable objects

A comparison counts verified differences, pricing differences (a differing free-tier or business-model fact, plus the entry price when both are verified on the same basis), and each side’s verified changes in 30 days. It never names a winner. Current verified offers are shown in their own block and never enter a count. A comparison is indexed only when both tools carry at least four comparable verified facts and differ on at least two.

How people use ToolMarketCap (behavioral data)

ToolMarketCap keeps first-party, aggregate daily counts: views of public objects (a comparison, a published stack or collection, a weekly pulse), arrivals through shared links (which kind of object, which channel — never who shared it), Switch explorations by reason, monitor actions by page, and clones. A count has no visitor identifier, IP address, user agent, or session attached. A browser counts each object at most once per tab session; nothing is counted under Do Not Track, Global Privacy Control, or browser automation, and the server refuses recognizable crawlers and link-preview fetchers.

None of this is shown publicly yet. “Most compared on ToolMarketCap” would need at least 500 comparison views in seven days and five pairs with 25 or more each; aggregate Switch reasons for a tool would need 100 explorations in 30 days, and a reason is shown on its own only with 20 or more (smaller groups are folded together). Until then these remain private operational counts. They are never inputs to signals, heat, alternatives, comparisons, or any ordering, and they describe use of ToolMarketCap, never the quality of a product. The owner of a published stack sees its own views and clones; nobody else does.

Tool pricing pages

Each tool has a pricing page with its verified plans, every observed price, any list-price movement, and the receipts behind them. A pricing page is indexed only when it holds at least two verified plans or a price observed more than once; thinner pages still open but are not promoted. Competitors’ entry prices are ordered by price, which is not a ranking.

How prices are labelled

A price is shown exactly as the vendor publishes it, in its own currency — never converted. A monthly price is billed monthly. An annual price is the per-month rate the vendor states for annual billing, and is always labelled annual; it is never folded into a monthly figure. Per-seat and per-user prices say so; usage prices name their unit (per resolution, per minute). Introductory or promotional prices are labelled as such and never become a tool’s entry price. Only a monthly, non-promotional list price in US dollars counts as an entry price.

Commercial policy and verified offers

Commercial relationships never influence the record. No vendor can buy a verification, an Activity value, a place in any ordering, a comparison outcome, an alternative listing, or an editorial conclusion. A tool with an affiliate programme gets no advantage over a tool without one; nothing organic on the site reads commercial configuration at all.

A verified offer is an offer the vendor publishes on its own site — for students, startups, non-profits, or for a limited time — read on that page and shown with its conditions and a receipt. It never replaces the verified list price. An offer leaves every page automatically once its stated end date passes. Offers are listed by tool name, never by discount.

“Verified offer”, “ToolMarketCap exclusive” (available only through ToolMarketCap), and “Affiliate link” (we may earn a commission) are independent labels, and each is shown wherever it applies. Today no listed offer is exclusive or an affiliate link: ToolMarketCap has no commercial relationship with any vendor. Should one exist, the offer says so and links here: we may earn a commission; this does not affect verification, rankings or intelligence.

Vendors can ask to represent their tool. An approved representative may suggest updates and offers through the same review as anyone else; it gains no ability to change the record directly.

Email, follows, feeds, and analytics

Subscription. Subscribing is not anonymous: ToolMarketCap stores your email address, the page you subscribed from, when you confirmed, and your email preferences — only to send what you asked for. Nothing is sent before you confirm through the link we email you, and every message has a one-click unsubscribe. After an unsubscribe the address is kept only as a suppression record, so it cannot be re-added without a new confirmation. There is no account and no password.

Follows. The tools and segments you follow are stored with your subscription, by their canonical record, to build your digest and your Following page. Opening a link from one of our emails signs that browser in with an HttpOnly cookie that holds a signed token, not your address; “Sign out of this browser” removes it.

Saved views and your stack. A saved Screener screen or Switch result is stored with your subscription as its title and its URL settings. Your stack holds the tools you add and, only if you type them, seats and a monthly spend. Both are private to you, never shown to anyone else, never used by any public page or ordering. You can delete a view or remove a tool at any time.

Delivery records. For each email we keep when it was attempted and sent and the provider’s message id — not the message itself. Email is sent through a transactional email provider, which necessarily receives your address and the message.

The RSS feeds of changes and briefs carry the same record with no address at all.

Product analytics are optional and off unless configured. When on, they record which kinds of pages are opened, which share buttons are used, and how search is used: the search text and the result you open, but only for searches that returned results and do not look like an email address, phone number, link, or credential — those are counted without their text. Events carry an anonymous identifier that exists only for the current browser tab. ToolMarketCap does not set cookies for analytics, does not create person profiles, and does not attach a name, email address, subscription, or follow list to any event; a click from a digest email is counted only as “came from a digest”. Nothing is sent when a browser signals Do Not Track or Global Privacy Control.

Like any request your browser makes to a third party, an analytics request reaches the provider with its network details, including your connection’s IP address. ToolMarketCap cannot prevent that from the page itself, so it requires the provider’s “Discard client IP data” project setting to be enabled before analytics is turned on in production. That setting is the control; this page does not promise more than the provider enforces.

Corrections and suggested updates

Anyone — a vendor, a customer, a researcher — can suggest an update. A suggestion is an input to our research, never a fact on its own: the sender verifies their email address, an editor reviews the claim, and nothing reaches the record until it is confirmed and recorded with a source and date like every other fact.

Where the claim can be checked on the vendor’s own public page, the record cites that page, not the person who pointed us to it. A vendor statement that cannot be checked publicly is recorded as vendor-reported and is never shown as verified. A company email address helps us weigh who is speaking, but it is not a verification. We never publish a sender’s address or the text they submit, and we don’t fetch submitted links automatically.

Coverage

ToolMarketCap tracks professional AI tools that are current and publish first-party pages it can read. A tool whose own pages say it is being discontinued, or whose last public release is more than a year old, is not added. A tool whose pages cannot be read without a browser may be tracked with fewer facts, and everything it lacks shows as “Not yet verified.” Other names a tool is known by — such as a name it had before a rebrand — are kept only for search, and only when the vendor’s own page or redirect shows them.