How ClickTug counts clicks

What counts as a click

A real user gesture on an outbound link, on a tool's own ClickTug listing page, reported by the visitor's browser.

Clicks on category cards, hub pages and comparison tables are recorded separately and arenot included in any published figure. That is deliberate: using one surface for both the numerator and the denominator is the only way published clicks divided by published views is genuinely the published click-through rate, rather than two numbers from different pages divided by each other.

What counts as a view

A listing page rendered in a visible browser tab for at least 1.5 seconds. Pages your browser loaded in the background because you hovered a link, or because a search engine prerendered a result, are not views.

What we exclude

Two kinds of rule. The first are applied when an event arrives: certain, cheap, and their worst case is one lost click. The second are applied when the daily figures are computed, because they need context beyond a single request or have a threshold that will need tuning.

Nothing is ever deleted for looking automated. Events are marked, never removed, so that a rule change can be applied backwards and so that every listing can publish how many clicks were excluded alongside how many were counted.

Applied when the event arrives

• Known crawler and automation clientsRequests identifying as a bot, crawler, spider, headless browser, HTTP library or command line client are refused before anything is stored.
• Requests that did not come from a ClickTug pageA report must arrive with a matching origin and the same-origin fetch metadata a browser sends automatically. A scripted client almost never reproduces all of it correctly, and a browser cannot get it wrong.
• Reports without a valid page signatureEvery trackable link carries a signature computed when the page was built. A report whose signature does not verify is refused. This stops a loop of forged requests. It does not stop a determined attacker, because the signature is visible in the page source, and we are not going to claim otherwise.
• Repeat clicks by the same visitor within 15 minutesOne counted click per visitor per tool per 15-minute window. The window is fixed rather than rolling, so a visitor clicking at 14:59 and 15:01 is counted twice while one clicking at 14:01 and 14:59 is counted once. Repeats are recorded as duplicates and reported in the excluded total.
• Prerendered and prefetched page loadsA page view is only recorded once the page has been in a visible tab for 1.5 seconds. Browsers load search results and hovered links in the background; counting those would inflate views and quietly deflate every click-through rate on this site.

Applied when the figures are computed

• Clicks with no matching page viewA click reported from a page load that reported no view of the same listing. Requires joining two tables across two requests, so it is decided after the fact.
• Visitors clicking more than 15 different tools in a dayNeeds the whole day to judge. A person cannot be classified on their third click, and guessing early would exclude genuinely curious readers.
• A single network accounting for over 40 percent of a tool's clicksApplied only on days where the tool had more than 20 clicks. Needs the day's whole distribution.

What we deliberately do not exclude

VPN and datacentre traffic is counted. A large share of genuine visitors to adult AI tools use a commercial VPN, and commercial VPN exit points live on datacentre networks. Dropping that traffic would look rigorous and would remove a substantial number of real people from the count. We record the network instead and judge concentration across a whole day.

Partner-link clicks are counted. Some outbound links carry a partner parameter and are marked on the listing that carries them. Those clicks are counted exactly like every other click. Excluding them would be a weaker claim than this one: money does not move the number, in either direction.

What we cannot measure

Clicks are counted in the browser on a real user gesture. A visitor with JavaScript disabled follows the same link and is not counted. There is no server-side redirect on this site to catch them, because a counting URL sitting in the page source is fetched by every link checker, unfurler and crawler on the internet, and filtering that would be a permanent arms race against the one number this business depends on.

Every published figure is therefore a floor, not a ceiling. We never estimate to close the gap. The moment one number here is modelled, every number here is.

When a figure is published

A number produced from too small a sample is worse than no number, so listings pass through three states. Counting began on 8 Aug 2026.

A rank is never published as a bare position. "Third of forty-one ranked image generators" is a fact; "third" on its own is not. A category with fewer than 5qualifying tools is not presented as a ranking at all: it is ordered by editorial review and the page says so.

How pages are ordered before then

Every hub on this site currently says "Ordered by editorial review", and that sentence deserves more than a link. Here is exactly what produces the order you see.

Each listing file carries one whole number called order. The hub sorts on that number and breaks ties alphabetically by product name. Nothing else touches the sequence. The number is assigned by hand when the listing is written and edited by hand afterwards, and no code anywhere on this site computes it, weights it, or derives it from a field. Any listing's position is checkable in our own repository, because the number lives in the same file as the facts.

It is a position, not a rating. There is no ClickTug score, and that statement has to survive contact with this page or it is worth nothing. So: theorder value is not a quality figure, is not published as one, is not aggregated, is not emitted in our structured data, and is not the sum of anything. A listing three places above another is not three units better than it. If we ever build a scored ranking, the rubric will be published and versioned before the first number appears, which is the same commitment made on the about page.

Partner status is not an input. A listing whose outbound link carries a partner parameter is ordered exactly as it would be without one, and its clicks are counted exactly as everyone else's are. That is enforced by the sort itself: the field the hub reads has no relationship to the affiliate field, and neither the click pipeline nor the threshold calculation can see it.

What screening finds

Candidates go through an automated screen before anyone spends time on them: does the vendor publish a price, terms, a privacy policy and a position on minors, each at an address a reader can open. The screen is reproducible from our own repository and its raw output is kept. It decides what is worth reading, never what gets listed.

On 8 Aug 2026 we ran it over 405 candidate services.267 returned anything readable to an automated request. Of those:

57 vendors published a price, terms and a privacy policy. 47 of them, 82 percent, failed on nothing except minors.

This market is not undocumented. Around half of it publishes terms and a privacy policy, which is what any small software company does. It has collectively decided to say nothing about minors, and until 9 August 2026 we required it.

Why we stopped requiring a minor safety policy

The number above was collected to justify the rule and ended up retiring it. A requirement that removes four out of five otherwise-complete vendors is either finding something real or measuring the wrong thing, and three findings from our own work say it was the second.

A clothing-removal service in that 31 publishes a minor safety policy, and we will not list it at any standard. A companion app we excluded on independent testing published one as well, along with a price, terms and a privacy policy: complete paperwork, excluded on the record. And two of the better-run services we assessed publish nothing on the subject anywhere, with no adverse finding against either.

A policy page is a claim. The record is evidence. So the requirement became a research step. Before anything is listed we search for reporting, incident-database entries and independent testing naming the vendor, and evidence that a product has generated sexual content involving minors, or people presented as minors, excludes it whatever its terms say. That check has excluded vendors whose documents were complete, which the old rule could not do and the new one has already done.

What a vendor publishes about minors did not stop mattering. It moved: every listing carries a row reading either Published, linked, or Not published with the date we looked and where. A reader now sees the silence instead of us acting on it out of sight. We still record who publishes one on our assessed but not listed page, and the five entries the change made obsolete were withdrawn there rather than left standing.

Freshness

Figures are recomputed and republished on a schedule, and every listing states the cutoff of its own data. A figure on this site is a statement about a period that ended on a date, not a live counter. That is deliberate: a number that changes when you reload is harder to quote and harder to check than one with a date attached.

How to audit us

If you operate a listed tool, your own analytics should show referrals from clicktug.com carrying a ref=clicktug parameter. We link to your real URL rather than through a redirect, and we do not suppress the referrer, precisely so that this check is possible.

If your number and ours differ by more than twenty percent, write to[email protected] and we will publish the reconciliation.

Privacy

The measurement described on this page sets no cookie, stores no identifier on your device, and never stores a raw IP address. The visitor value used to deduplicate clicks is a one-way hash salted with a secret that changes daily, so it cannot be linked across days or reversed. Separately, this site runs Google Analytics for visit counts, which does set two cookies and plays no part in any figure published here. Theprivacy policy sets both out in full, including how to opt out.

Changes to these rules

When an exclusion rule changes, affected figures are restated and the change is recorded with its date. Rule changes are the reason excluded events are marked rather than deleted.

• 7 Aug 2026Version 1. Initial counting rules, as described on this page.