Analytics for builders

Five numbers to read in the week after launch

By Nora Blake · Nov 2, 2026 · 10 min read

A small storefront on its opening day with a few visitors at the door and a handwritten ledger on the counter, where each arrival is noted by name.

You launched, you open the dashboard, and the numbers are small. That is the normal state, and it makes most analytics screens misleading. Averages, rates and trend lines need hundreds of visits before they say anything. Counts and names do not. One real visitor from a named source is a fact. "42% bounce rate" on 19 visits is a rounding artifact.

So in week one, read counts and names, and leave rates for later. The five numbers below are the ones that still hold up at low volume, with a note on when each one lies to you.

Part of the problem is that a launch week is not only people. On the network Cloudflare measures, bots are close to half of what requests HTML pages, and a tiny site sees that mix with no cushion of volume to hide it.

47%44%HumanNon-AI bots

Source: Cloudflare, The 2025 Cloudflare Radar Year in Review, Dec 2025. Retrieved Oct 30, 2026.

That chart is network wide, from one company's traffic, for HTML requests only. It does not describe your site, and Cloudflare notes that its HTML-only figures differ from what Radar shows for all content types. It is here to set an expectation: on a new site, some of what arrives is not a person, and the first job of your numbers is to separate the two.

1. People per day, with your own visits removed

The first number is how many different people came on each day. Look at it per day, not as one total for the week.

There are two reasons. First, the hash that the People number counts mixes in the date and is re-salted every day, so nobody can be followed from one day to the next. That is a privacy property, and it has a side effect you should know about: if you add up seven daily People numbers, you get person days, not people. Someone who read your page on Monday and again on Thursday is counted twice in that sum. The returning versus new split comes from a second hash without the date, which is a separate number and the one to use for that question.

Second, your own visits are the biggest distortion in week one. You reload the page after every change, your co-founder checks it, your agent runs a smoke test. If you launched with 30 real visitors and you made 40 visits yourself, the chart is mostly you. Remove them before you read anything. The post on keeping your own visits out of site analytics covers how.

Then check that the tracker works at all. If People is zero after you have visited from a second device, the problem is the install, not your launch. Install analytics and prove it works in one agent session is the fast way to rule that out. Peeko is one script tag with a site key, sets no cookie, and needs no configuration once it is on the page, so there is not much that can go wrong, but a script in the wrong template is the usual cause.

What to do with the number: compare day one with day two with day three, and ask only whether it is zero or not zero. A day with 6 people and a day with 11 are not different in any way you can act on.

2. The name of your top referrer

The second number is not a number. It is a name: the single source that sent the most people.

At low volume, percentages of referrers are useless. If 30 of your 40 visitors came from one post on a forum, the lesson is "that forum post worked", not "75% of traffic is social". Read the source by name, open it, and look at what was said there and what people clicked.

Direct traffic deserves a caution. A visit with no referrer is not necessarily someone typing your address. Links in chat apps, email clients and some browsers arrive with no referrer at all, so a large direct bucket in week one is mostly "unknown", and you should say that to yourself when you read it.

Assistants are a source too. Visits that arrive from ChatGPT, Claude, Perplexity, Copilot and Gemini are counted from ordinary page views by their referrer, with no middleware involved. If one of those names shows up in your first week, that is a person who clicked a link inside an answer. Count visits from AI assistants by referrer shows how that count is built and where it undercounts.

What to do with the name: write down the top three sources by name, and repeat what worked. Do not compare their shares.

3. Whether anything has fetched your pages yet

Here is a question the visitor count cannot answer: has anything fetched your pages yet? A count of zero on day two is a real answer, not a bug, and it is more useful than it looks.

Crawlers do not run JavaScript, so a script tag never sees them. That is why AI crawler tracking in Peeko is a separate server side path. The @usepeeko/bot-track middleware is ESM only, needs Node 20 or newer, has zero runtime dependencies and weighs under 3 KB gzipped. Peeko matches the user agent against a curated list of 55 agents from 21 companies and files every recognized hit under AI answers, Search indexing, AI training or Other AI bots.

Those four labels matter because they mean different things. A hit under AI training says a company fetched your page to learn from it. A hit under AI answers says a page was read while an assistant was replying to somebody. The question itself never reaches your server, so nobody can report it, and the count is all you get. Search indexing is the one you want first, because a page that was never fetched for a search index will not show up in search at all.

On privacy, the crawler path stores no address and no visitor hash at all. It keeps the raw user agent, up to 512 characters, so old hits can be reclassified when the list grows. And AI crawler hits are never billed.

What to do with the number: if it is zero and the middleware is installed, wait and check again on day four. If it is zero and the middleware is not installed, the zero means nothing. AI crawlers on a new site: the first 30 days describes what a first month tends to look like, so you know what normal is before you worry.

4. The count of one click or one goal

The fourth number is how many people did the one thing you launched for. It can be small. Three is a valid answer.

You do not need to write code to start. Five kinds of click are recorded with no code at all: outbound, phone, email, whatsapp and form. If your launch goal is "someone fills in the contact form" or "someone clicks through to the store", it is likely already being recorded. Anything else is a data-peeko-event attribute on an element, or a call from your own script:

<button data-peeko-event="signup-click">Start</button>
window.peeko.event("pricing-viewed");

An event takes one name and one optional string, and there is no properties object, which keeps the data small enough to read at a glance.

A goal is a definition rather than a counter. A goal written today counts over months before it was written, and a site can hold twenty of them. That removes the pressure to guess the right goal on launch day. Launch, look at what people clicked, then write the goal, and it will already hold its history. If your count is tiny, your first goal with almost no conversions is about reading a goal when the count is 2 or 3.

What to do with the number: do not compute a conversion rate. Three conversions out of 40 visitors is 7.5% today and 5% if one of those conversions turns out to be you, and neither figure predicts next week. Read the three and open the sessions or sources behind them.

5. Search impressions, read late

The fifth number is how many times your pages appeared in search results. It is the only number here that tells you about people who have not found you yet, and it is also the one that arrives last.

Google's own help page says collected data should normally be available in two to three days, and that it can take up to a week to generate data for a newly created site or one newly added to Search Console. Peeko's Google Search Console and Bing Webmaster Tools connections are read only, use encrypted tokens, and keep their rows for sixteen months. Google publishes search data about three days late and revises its three most recent days, so a low number on the newest days is not a real drop. Bing's query rows are weekly buckets dated to a Friday, which is why a Bing position is derived and can be null.

That gives a clear rule for week one. Do not read search on day one or two. On day five or six, ask a yes or no question: did any query show impressions at all? If yes, note the query, because it shows what people think your page is about, which may differ from what you think. Position is noise at this stage. An average position built from 4 impressions can swing from page one to page six when one more impression is added.

Numbers to ignore for now

Bounce rate. Plausible's documentation defines it as the percentage of visitors who land on a page and leave without meaningfully engaging with your site. On a launch day with 25 visitors, that is 25 yes or no answers, and most of them came from a link someone shared. It needs volume, and it also depends on what each tool counts as engagement, so a number from one tool is not comparable to a number from another.

Visit duration. The same documentation defines it as the amount of time visitors spend on your site. A handful of long reads from people who left a tab open will dominate an average of ten.

Country and device splits. At 40 visits, a country is one person. These become meaningful at a few thousand visits.

Growth percentages. Going from 3 to 9 is 200% growth and nothing else. Use the counts.

Anything that includes your own traffic. Fix that first, as in number 1.

A routine that fits in ten minutes

You do not need a daily ritual in week one. Two short checks do the job.

  • Day 1 or 2: Confirm the tracker works from a second device. Confirm your own visits are excluded. Write down your top three sources by name.
  • Day 5 to 7: Read People per day as zero or not zero. Read the crawler count by category. Read the count behind your one goal. Check whether search impressions have appeared.

The trial is fourteen days with no card, it starts when you create your account, and it covers one site. A launch week plus the week after fits inside it. If you work with an AI coding agent, Peeko hosts a read only MCP server, and the command line needs no install, since npx @usepeeko/cli speaks the same endpoint an agent does, so a command and a tool call return the same numbers. That means you can ask your agent for these five numbers instead of opening a dashboard.

The point of week one is not to learn how your site performs. It is to find out whether anyone and anything arrived, from where, and whether one thing happened. Five counts and names answer that. Everything else can wait for the traffic to carry it.