← All resources
Guides
The problems people arrive with
Short answers to the questions that bring teams here, each with the mechanism rather than the marketing version.
How to separate bot traffic from real visitors
- Why the usual approach fails
- User-agent filtering catches the bots that identify themselves and nothing else. A list maintained in a config file ages badly, and the traffic that most distorts a report is the traffic least interested in announcing itself.
- What works better
- Classification at the edge, where the request arrives with its network context, reputation signals and behavioural pattern — not just the string it claims to be. Verified crawlers are confirmed rather than trusted.
- The result worth having
- A default business view where 'people' means people. Every other category stays available, because AI crawler volume on your documentation is genuinely useful information — it is just not sales.
How to track ChatGPT and AI referral traffic
- The distinction nobody makes
- An AI crawler reading your site is a bot. A person arriving from an AI assistant that recommended you is a customer. They have opposite commercial meanings, and most analytics tools file both under 'referral' or miss both entirely.
- How the referral is observed
- Through referrer, campaign parameters and integration data where the assistant makes it observable. Coverage depends on what the assistant sends, which varies — so the honest version reports what was seen rather than estimating what was not.
- Why it is worth the trouble
- Assistant traffic usually arrives in small volumes, and its conversion rate often differs from the rest of referral traffic — which you can only see once it is separated out. Averaged into 'referral' it disappears entirely.
How to know why conversion rate changed
- Start by decomposing, not explaining
- A sitewide number moving 19% almost never means everything moved 19%. Break it by device, browser, channel, landing page and country first — the answer is usually one segment doing something dramatic.
- Then rule things out
- Did traffic volume change? Did the acquisition mix shift? Did errors rise? Did a release go out? Each one you eliminate makes the remaining explanation stronger.
- Then be careful with the word 'because'
- A release at 14:02 and a decline from 14:20 is a strong correlation on a short timeline. It is not proof, and treating it as proof is how teams roll back the wrong thing.
How to connect website analytics to revenue
- Browser conversions are not enough
- A click on a signup button is an intention. Whether the trial started, the payment cleared and the refund arrived three weeks later are facts your server and billing system hold, and only they know them.
- Send server events
- POST the outcome with a pseudonymous customer key, a value, a currency and an idempotency key. The idempotency key is the part people skip and the part that matters — a retried webhook is a duplicated sale.
- Then read the money from where it lives
- Connect the billing, commerce or finance system rather than recomputing revenue from events. A figure that reconciles with what finance already believes is worth more than a more precise one that does not.
How to measure revenue per landing page
- Rank by contribution, not traffic
- The page with the most visitors is frequently near the bottom by revenue per visitor. Both numbers are true and only one of them tells you where to spend next week.
- Attribute honestly
- A landing page rarely closes a sale on its own. Pick an attribution model deliberately, show which one produced the number, and treat the result as model output rather than as a fact about causation.
- Go below the URL when it matters
- With content assets connected, the same headline across four pages is one measurable thing instead of four weak rows — which is usually where the real difference lives.
How to measure deployment impact on conversion
- Annotate releases
- Commit, release ID, environment, timestamp and changed routes. Without the marker you are eyeballing a chart and remembering wrong.
- Compare like for like
- Same weekday, same hours, same segments. A Tuesday-afternoon release compared against a weekend average will show an effect whether or not one exists.
- Watch the technical measures alongside
- If responsiveness moved in the same segment and traffic did not, the technical explanation is worth investigating first. That is still a correlation — but it is a correlation with a mechanism.
Keep reading
The rest of the documentation
Connect a site today. Read tomorrow's brief instead of building it.
Connect a site and the first brief arrives with the day's changes already explained — traffic separated from bots, conversions attached to revenue, and the evidence behind every sentence one click away.
Real people separated from bots Every answer shows its evidence Reveliqo runs on Reveliqo