Traffic problem or first-screen problem? How to tell

Imagine one visitor who searched for exactly what you make and read for three minutes, and fifty who tapped a link in a social feed and left after four seconds. The average visit comes out at a little under eight seconds. Nobody had that visit. The one person you served well disappears into the people you did not, and the fix for each group is different.

So the first step is not a new metric, it is a split: the same engagement numbers, broken down by where the visit came from.

One note on definitions before comparing anything. Tools measure engagement differently. In Google Analytics 4, an engaged session is one that lasts longer than ten seconds, includes a key event, or has at least two page views, and bounce rate is simply the share of sessions that were not engaged. Other tools define bounce and time on page in their own ways. Check what your tool means before you compare sources, and compare sources within one tool rather than across tools.

Visitors from different sources are not just different amounts of interested. They arrive asking different questions.

Social arrivals have no frame. Someone who tapped your link on X, Reddit, LinkedIn or in a group chat was usually doing something else. Often they are on a phone, between two other posts, and all they know is the line of text around the link. Their question is “what is this?”. A first screen that opens with a slogan, a gradient and a logo gives them nothing to hold on to, and they go back to the feed.

Search and AI-referred arrivals have already named their problem. Someone who typed a query, or asked an AI assistant and was sent to you, arrives knowing what they want. Their question is “can it do my specific thing, and where does it break?”. They want the supported inputs, the limits, the price and how you compare. A vague hero costs them a scroll; hidden limits send them back to the next result.

Launch communities such as Product Hunt and Hacker News sit somewhere in between. Visitors arrive with the frame of whatever title was posted, and tend to look quickly for how it works and what it costs.

The same first screen cannot answer “what is this?” and “where does it break?” equally well. Most pages answer neither, which is why splitting by source is so informative.

In Google Analytics 4, the Traffic acquisition report shows sessions by channel, and you can switch the dimension to session source and medium. Look at engagement rate, average engagement time and your sign-up event for each row. Most other analytics tools have an equivalent sources view. Then read the pattern:

A caution about small numbers. If a source sent you thirty sessions, its engagement rate can swing wildly from one week to the next by chance alone. Treat differences between small groups as hints, look at the direction rather than the decimals, and back them up by watching what people actually do.

On a young site, “direct” is often the biggest source in the report, and it is the one most worth distrusting. Direct does not mean someone typed your address. It means the analytics tool received no information about where the visit came from. On a new site that bucket typically fills with:

• Links shared in messaging apps, email and many mobile apps, which often open without passing a referrer.
• Links from sites or apps whose referrer policy strips the source.
• Browsers and extensions that limit referrer information for privacy.
• Your own visits: testing the deploy, showing the site to a friend, checking a typo on your phone.
• Some visits from AI assistants. Some AI products tag their outbound links and some do not, and that changes over time, so look at which sources actually appear in your reports rather than assuming.

Two fixes. First, exclude yourself: GA4 can filter internal traffic by IP address, and most analytics tools have a similar setting or an opt-out for your own browser. Second, stop letting your own shares arrive untagged.

UTM parameters are labels you add to a link so the analytics tool knows where the visit came from even when the referrer is missing. Three are enough for most people:

https://example.com/?utm_source=reddit&utm_medium=social&utm_campaign=launch

• Pick a naming convention and keep it. Use lowercase throughout: in GA4, “Reddit” and “reddit” become two separate rows.
• Tag the links you share by hand too: newsletters, group chats, a message to a community you belong to. That is exactly the traffic that would otherwise vanish into direct.
• Never put UTM tags on links between pages of your own site. They overwrite how the visitor actually arrived.
• Keep a short list of the tagged links you have posted and when. It turns a spike in your report into something you can explain.

If the split shows that people without context leave, these changes help them without getting in the way of the visitors who arrived knowing what they want:

1. One plain sentence that says what the product does, for whom, with what goes in and what comes out. No metaphor, no slogan. If you need help getting there, the post on silent copy failure has a method.
2. Show the actual product. A real screenshot or a short clip of the real output, not an illustration. A screenshot is evidence; an illustration is decoration.
3. A way to see it work without signing up: sample data, an example output, a read-only demo. For visitors who give you a few seconds, this is often the only thing that can turn a glance into interest.
4. Limits near the top, for the search arrivals: what it supports, what it does not, and what it costs or that it is free while in beta.
5. Keep the promise of the link. If your post said “turns bank exports into accountant-ready spreadsheets”, the first words on the page should say that, not something new.
6. Look at the first screen at phone width. Social visitors are often on a phone. If all that fits above the fold is a logo, a gradient and half a headline, that is what they judged you on.

Then read the numbers by source again over a comparable period. A before-and-after comparison is not proof, because the mix of visitors changes from week to week, but a large shift in one source after a change aimed at that source is worth knowing about. The post on testing with little traffic goes into what a before-and-after can and cannot tell you.

This split, different first screens for visitors who arrive with different questions, is the problem IntentGrid is built around: it rewrites a page's copy for each visit based on the traffic source and other context available at page load. It is in early access, it produces no measurement, and it cannot fix a first screen that has nothing specific on it, because it only reshapes what is already there. How it works has the details. Do the fixes above first either way.