Platform How it works Live demo Agencies FAQ Contact Request early access

Your analytics counts visits. PageSight finds what is broken.

It scores every visit, names the reason people leave, puts a price on it, and ranks the fix by what it pays back.


PageSight is developed by Cygnik Tech on the website monitoring platform Cygnik already runs on its own sites, including this one. It is in private preview: the platform is built, and we are onboarding a small number of sites by hand.

Every morning, the three things worth your attention.

Written overnight from your own measured data. Each finding cites a number from your site and ends with something to do.

  • Never invented. The AI layer is instructed never to make up a metric, never to claim certainty when data is sparse, and to say so plainly when there is nothing to report.
  • Ask a follow-up in plain English. One click loads the finding's context into a chat; you decide whether to send it.
  • Nothing is applied for you. Every draft the AI produces waits for an explicit human confirmation.
Today/briefing
tracker live
This morningwritten overnight

Your checkout form loses most of its drop-off at the phone field, on phones. Test tap-target size and keyboard mode first.

Convert

One paid source is sending mostly automated traffic. Pause it before the next budget cycle and re-check the numbers you upload to the ad platforms.

Protect

Visitors arriving from AI assistants convert above your baseline. Worth more of your attention than their share of traffic suggests.

Reach
New error on /checkout First seen just now. Fingerprinted, so you hear about it once.
EmailSlackWebhook

Diagnostic questions, not descriptive charts.

Every screen is built around a question a business actually asks. Pick one.

Ad waste in currency, not percentages.

Every visit is scored for bot-likeness on dozens of independent signals, from automation flags and headless fingerprints to the smoothness of the mouse. Then the fake traffic is priced against the cost-per-click you set, so the number can be reproduced in front of a finance team.

  • The dirtiest source, named. Campaign sources ranked by dollars wasted, not raw bot count, so the first thing you pause is the thing costing the most.
  • A bot rate with a confidence interval and a three-way human, suspicious, bot split, instead of one unqualified percentage.
  • Scrapers caught by technique. Automation drivers, missing browser APIs, dishonest user agents, datacentre networks, and crawlers seen working across many PageSight sites at once.
  • Contaminated conversions flagged before you upload them to Google and Meta.
  • Honest instrumentation. If sessions exist but none carry a score, it says so rather than reporting a comforting zero.
Protect/authenticity
trailing week
Who is actually visiting
humansuspiciousbot
Spend lost to bots, by sourceworst source named
display-retarget
paid-social
search-generic
search-brand

A stop-reason for every lost sale.

Bounce rate is not a diagnosis. Every engaged visitor who did not buy is labelled with why: price shock, comparison shopping, information seeking, trust doubt, interrupted, technical failure, or decision fatigue. Each reason maps to specific interventions, several with a ready-made test.

  • See who is about to buy, live. The highest-intent sessions active right now, each with a probability you can open and read signal by signal. Nothing is a black box.
  • Dead clicks and rage clicks caught in the browser, including buttons that silently do nothing and never appear in any server log.
  • Forms diagnosed field by field. Time per field, re-edits, and the exact field people gave up on, without ever recording what they typed.
  • A/B tests that stop themselves the moment the evidence is decisive, with per-cohort results that show where the "winner" actually loses.
  • Errors priced by the conversions they cost, against a matched control group on the same pages, devices and countries.
Convert/exit reasons
/pricing
Why engaged visitors left
price shock
comparison shopping
trust doubt
technical failure
interrupted
decision fatigue
About to buy, right nowlive
Safari on iPhone, Toronto
/pricing to /security to /pricing
high intent
Chrome on Windows, Berlin
/docs/install to /pricing
high intent
Chrome on Android, Mumbai
/blog to /pricing
rising

Are AI assistants reading your site?

A question almost nobody can answer yet. PageSight shows real visits arriving from ChatGPT, Perplexity, Claude, Gemini, Copilot and You.com, how many converted, and which engine over-delivers relative to its share of traffic, so effort goes where it pays.

  • Dark social recovered. Arrivals from WhatsApp, Slack, Instagram and other in-app browsers that ordinary analytics writes off as "direct", with the pages people are actually sharing.
  • Articles worth refreshing, ranked by money. Content whose traffic has decayed below its peak, with an estimated revival value per article.
  • Which paragraphs people read and where they skim or abandon, measured from real dwell time on your own readers.
  • Send when your audience is awake. Peak day and hour fitted from your own activity curve, per social platform.
  • Audiences from behaviour, with no email list. Returning visitors, pricing-page viewers, form abandoners, AI-referred visitors.
Reach/AI referrals
trailing month
Visits arriving from AI assistantsvs. your baseline
ChatGPT
most visits
converts above baseline
Perplexity
fewer visits, high intent
over-delivers
Claude
growing
at baseline
Gemini
occasional
at baseline
Copilot
occasional
below baseline
Where "direct" traffic really came from
WhatsApp
Slack
Instagram app
LinkedIn app

Names, not numbers.

Visitors are grouped by the office network they came from and labelled with the organisation, so a count of sessions becomes a company, the people behind it, and the evaluation pages they read. Company identification is an inference from the public network registry, and it says so.

  • Spot a buying committee before sales does. Several roles from one company reading the same evaluation pages in a short window.
  • An early churn warning at account level, from falling traffic, people dropping off the account, comparison-page visits and rage clicks.
  • Accounts that are quietly growing, and word-of-mouth champions detected from share patterns rather than personal data.
  • A case file per visitor: every visit, their colleagues, buying intent over time, and the page that moved them.
  • Privacy first. Emails are hashed in the visitor's own browser before anything is sent. Raw IP addresses are never stored.
Engage/accounts
this week
Active buying committeewarming up
Logistics company, Toronto network
several people, evaluation pages this week
call first
Evaluator
/pricing to /integrations to /security
inferred role
Engineer
/docs/install to /docs/privacy
inferred role
Decision-maker
/pricing to /contact
inferred role
Accounts by lifecycle stage
visitorevaluatorleadcustomerat risk

Every page, valued like a property you own.

One picture of the whole site: every page sized by real human traffic and coloured by what it is worth. Value comes from a removal-effect model, "do visitors who touch this page convert more than those who don't", not from crediting whatever page happened to be last.

  • High-traffic dead zones. Pages that look like winners on a pageview report and contribute nothing.
  • Hidden gems. Small pages carrying the business that every traffic-ranked report buries.
  • One ranked repair list for the whole site, drawn from every detector, with an expected value, effort in hours and a confidence grade on each fix.
  • A health score with its workings shown. The pillars, their scores and their weights, never a bare number.
  • Dead ends, trends and a forward projection per page, with a confidence band that widens when there is less history behind it.
Pages/portfolio map
sized by human traffic
earninghidden gemneutralhigh-traffic dead zone
Repair list, ranked by value per hour of work
Fix the phone field on the checkout form
Convert · under an hour · high confidence
first
Pause the display-retarget source
Protect · minutes · high confidence
next
Refresh the decayed roundup article
Reach · half a day · medium confidence
later

One script tag. Then the platform does the heavy thinking on a timer.

Data moves through the platform in stages. Dashboards read pre-computed answers, which is why they open fast, and why the platform can reach out to you without being asked.

Capture

Notice what happened

A small script watches what real visitors experience: slow pages, dead buttons, hesitation, rage clicks, where reading stops. Optional edge helpers see the traffic that never runs JavaScript at all.

Collect

Check it, then file it

Everything arrives at one guarded front door. Identity, quota, origin, size, signature and a robot check, then enrichment with what only the network can know: country, provider, device.

Process

Score overnight

Timers do the expensive work on a schedule: scoring traffic, detecting problems, pricing them, refreshing baselines. Page outages are checked every few minutes against the same hour of the same weekday.

Surface

Make it readable

Screens organised by question, not by metric. Every card can be opened to see its workings, and any card can be explained in plain English.

Act

Tell someone

Alerts to email, Slack or your own webhook. A written morning briefing. Scheduled reports. And session replay for the moment a number says something went wrong.

This page is running PageSight. Try to break the box.

Press the button that does nothing. Click it in frustration. Hover the link and walk away. Start typing, then give up. The panel on the right is worked out in your own browser, live, from what you actually do.

Checkout

A deliberately broken form, for you to poke at.

Scroll this box. PageSight records how far a reader actually gets, and what was on screen at the moment they stopped, which is how the answer becomes "they leave at the newsletter interstitial" rather than "they leave at sixty percent".

On a real site the same measurement runs on every page, at every scroll milestone, for every visitor. It costs nothing to collect and it turns a vague bounce number into a specific place on a specific page.

The tracker also watches whether a click did anything at all. A button wired to nothing sends no request to your server, so it appears in no server log, in no error tracker, and in no analytics report. It is invisible until something is watching the browser itself.

That is the whole idea. Measure what the visitor experienced, not what the server delivered, then work out what it cost you.

What PageSight just noticedlive
  • Dead clickwaiting
  • Rage clickwaiting
  • Hesitationwaiting
  • Field abandonedwaiting
  • Broken call to actionwaiting
  • Reading depthwaiting
  • Connectionchecking
  • Device tierchecking
  • Arrived fromchecking
Worked out in your browser.

Find out before your customer does.

The platform contacts you. You do not have to remember to look.

  • A brand-new bug reaches you within minutes, without opening a dashboard.
  • Told once, not thousands of times. Alerts are keyed on the error's fingerprint. Mute one for an hour, a day, a week or forever.
  • Route to the tooling you already use. Email, Slack, or a signed webhook into PagerDuty, OpsGenie, Discord or your own endpoint.
  • "No data" means no visitors, not a broken pipeline. A synthetic visit is pushed through the real collector every few minutes, and alerts escalate sensibly during a long incident instead of repeating.
  • Page outages judged against your own rhythm. A quiet Sunday night is not an alarm; the same drop on a weekday afternoon is.
Settings/alerts
owner
Delivery channelstest send passed
Email
ops and two more
enabled
Slack
#site-incidents
enabled
Webhook
PagerDuty · HMAC signed
enabled
Recent alerts
TypeError in checkout.js
first seen this morning · release after deploy
open
Failed request to /api/quote
first seen yesterday
muted for a week
Traffic to /pricing far below its usual for a weekday afternoon
resolved itself
resolved

When a number says something broke, watch it happen.

Session replay records a privacy-masked, playable reconstruction of what a sampled visitor saw and did. Not video. The page's structure and the changes to it.

  • Triage ranked worst-first, with the moment of failure printed so you jump straight to it instead of scrubbing a long recording.
  • Frustration flagged in words. "Clicked the same thing repeatedly at this point. It probably looked clickable but wasn't working."
  • Verdicts on money pages. Was interested, didn't convert. Bounced off a key page. Looked around, didn't find it. Or, honestly, nothing notable.
  • Masked by default, not by configuration. Every input blanked. Password and card fields never leave the browser. Sampling and retention are two dials you control.
Replay/triage
worst first
Something broke early on, and they left from /checkout
Safari on iPhone · Toronto · shorter than your typical visit
error
Clicked the same thing repeatedly on /pricing
Chrome on Windows · Berlin · about typical length
frustration
Was interested, didn't convert
Chrome on Android · Mumbai · longer than typical
near miss
Looked around, didn't find it
Firefox on macOS · London · many short pages
hunting

Every number is checkable. Every privacy promise is enforced by machinery, not policy.

Unique visitors with nothing to leak

Counts are built from a one-way daily fingerprint that changes at midnight and never contains an IP address.

Measurement is off until you switch it on

The deeper capture features are individually switchable, all off by default, owner-only, and every flip is audit-logged. Switching one off takes effect within about half a minute.

Isolation enforced by the database

Per-row rules keep one customer's data invisible to another, and an inspection routine fails the build if a table is missing one.

Every signal documented in plain English

Each captured field lists its source, how often it fires, whether it may contain personal data, and its honest limitations. Filter the catalogue down to potentially-personal fields in one click.

"Delete my data", end to end, with proof

See the affected row counts before committing, confirm by retyping the identifier, then a window in which one click cancels it.

Emails hashed before they leave the browser

Lead capture converts an address to an irreversible fingerprint inside the visitor's browser. The servers never receive it. Typed text in forms is never recorded.

The questions you currently ask a stack of tools.

Each of these is usually a separate subscription and a separate tab. PageSight answers them from one set of data, and connects them: an error is priced by the conversions it cost, a slow page is tied to the device tier it is slow for.

Did something just break?
Sentry, BetterStack
Fingerprinted error alerts, page-outage checks, a self-testing pipeline
Where do people struggle on the page?
Hotjar
Rage and dead clicks, scroll milestones with what was on screen, replay
Why does the form lose people?
Typeform analytics
Time per field, re-edits, the field they gave up on, a named cause
Is the traffic even real?
a separate bot-detection tool
Bot scoring on every visit, wasted spend by source
Which content is decaying, and where do AI assistants send people?
Ahrefs, plus guesswork
Article decay ranked by value, AI-assistant referrals with conversion
Which companies are looking at us?
HubSpot, Intercom
Accounts by network, buying committees, churn and expansion signals
Which test won?
an A/B testing suite
No-code experiments that stop themselves at a decisive result, per cohort

What PageSight does not do.

Every module in the platform carries a scope fence, written from the code. Here is the short version, so nothing on this page is discovered by a customer rather than by us.

It does not block bots. It measures and prices them. There is no firewall, no CAPTCHA, no IP ban. Bot mitigation is advice, not enforcement.
Dollar figures are model outputs. They are built from the conversion values and cost-per-click you configure, not read from your payment system. Until you set those, the figures read zero, and the product tells you which field unlocks which number.
Company identification is inference. It comes from the public network registry, never from a paid enrichment vendor. It never claims to know who a person is.
It does not send anything on your behalf. No emails to your visitors, no outreach, no CRM sync. Drafts wait for your review.
It is not real-time. Most scoring runs overnight on a rolling window. Outage checks and error alerts run every few minutes.
Some detectors need traffic volume. A very low-traffic site will see honest empty states rather than invented findings. We say so up front.

Questions people ask before the first call.

What is PageSight?

PageSight is website diagnostics. Ordinary web analytics counts what happened: how many visitors, from where, on which page. PageSight answers diagnostic questions instead. It scores every visit, names what is wrong, prices it against the conversion values you configure, and ranks the fixes by what they pay back. It is developed by Cygnik Tech and is in private preview.

How is PageSight different from Google Analytics?

Google Analytics is a measurement tool that counts events. PageSight is a diagnostic tool that decides whether something is wrong. Where an analytics report gives you a bounce rate, PageSight gives a stop-reason for every lost sale: price shock, comparison shopping, information seeking, trust doubt, interrupted, technical failure or decision fatigue.

It also catches things a page-view counter structurally cannot, such as dead clicks, which are buttons that silently do nothing and therefore send no server request at all.

Does PageSight replace Sentry, Hotjar, or a bot-detection tool?

It answers the same questions from one set of data, and it connects them, so an error is priced by the conversions it cost and a slow page is tied to the device tier it is slow for.

It is not a like-for-like swap. A developer who lives in stack traces all day will still prefer a dedicated error tracker for raw exception triage, and PageSight does not block bots at all.

Can I buy PageSight yet?

Not yet. PageSight is in private preview. The platform is built and running, and Cygnik Tech is onboarding a small number of sites by hand. There is no public pricing and no self-serve signup. Access starts with a conversation through the contact page.

Does PageSight block bots?

No. PageSight measures bot traffic and prices what it costs you in wasted ad spend, ranked by campaign source. There is no firewall rule, no CAPTCHA and no IP ban anywhere in the product. Bot mitigation is advice you act on, not enforcement the platform performs.

Where do PageSight's dollar figures come from?

They are model outputs, not measured revenue. Every figure is built from the conversion values and cost-per-click that you configure. PageSight does not read your payment system, your order table or an advertising invoice. Until you set those values the figures read zero, and the product tells you which field unlocks which number.

How does PageSight protect visitor privacy?

Visitor counts are built from a one-way daily fingerprint that changes at midnight and never contains an IP address. Email addresses captured by forms are hashed inside the visitor's own browser before anything is sent, so the servers never receive them, and typed form text is never recorded.

Deeper capture features are individually switchable and off by default. Tenant isolation is enforced by per-row database rules rather than application code. Session replay masks every input by default, and password and card fields never leave the browser.

Can PageSight tell me which companies visited my site?

It groups visitors by the office network they came from and labels that network with an organisation name, so a count of sessions becomes a company and the people behind it. This is an inference from the public network registry, not a paid identity vendor, and PageSight never claims to know who an individual person is.

What does PageSight need in order to work?

One script tag on your site. Optional edge helpers can additionally see traffic that never runs JavaScript, such as scrapers and API clients. To turn on the dollar figures you also tell PageSight what a conversion is worth.

Some detectors need traffic volume to produce a result, so a very low-traffic site will see honest empty states rather than invented findings.

Can agencies run PageSight for their clients?

Yes. Sites are grouped under a client account, reports carry your own logo and colours, client staff get access scoped to only their own sites, and you can share a report by a link with an expiry and a view cap that you can revoke instantly. There is more on the agencies page.

Who builds PageSight?

Cygnik Tech, an IT and AI consulting practice based in Toronto, Ontario, Canada. PageSight is the commercial product built on the website monitoring platform Cygnik runs on its own properties.

Running this for clients?

Group sites under each client, put your own brand on the reports, and share them by a link that expires when you say so.

Bring your site to the preview.

Tell us where to reach you and a little about your site. We install the tracker with you and walk you through your first briefing.

Request early access