Process

Surfer — Standard Plan Onboarding

May 2026
Krystian Prorok

Starting point

0
Existing steps

No onboarding existed before this. Users landed in the product with no setup guidance.

0
Baseline data

No completion rates, no drop-off points, no time-on-step. All targets are forward-looking hypotheses.

The constraint

Design for a user who just paid. They expect the product to be smart. The onboarding has to feel worthy of that expectation.

Key decisions

1

Step-by-step flow instead of a single page

Alternatives considered

  • One long page with all setup sections visible at once

What was chosen

A sequential, step-by-step flow where each screen has one job.

Why

The steps are not independent — brand setup must come first because Pages and Prompts both depend on it. A single page would hide that dependency and leave users unsure of what to fill in first. The step order communicates that the setup has a logic: understand the brand, then configure what to track based on it.

Risk

More screens means more navigation. If any single step causes drop-off, the user loses all progress on the steps that follow.

2

Fast Pass — AI picks, user confirms

Full approach exploration

Alternatives considered

  • Show all pages ranked by impact score (Approach A)
  • Group by topic, user picks clusters (Approach B)
  • Full manual selection with no pre-selection (Approach D)

What was chosen

AI pre-selects the top 50 pages silently. User reviews but doesn't need to. One click to continue.

Why

At the start of onboarding, users don't yet know which pages matter. The AI has more signal than they do. Asking them to pick from 100 articles would stall the flow and erode confidence before the product has delivered any value. Low trust is earned through results, not through making users do work.

Risk

Bad AI pre-selection is invisible. If the quality is poor, the user won't know until they see wrong data downstream — at which point they blame the product, not the onboarding.

3

GSC-first, manual as secondary

Alternatives considered

  • Manual URL entry as the primary path
  • Both options presented equally side-by-side

What was chosen

GSC connection is the primary recommended path. Manual URL entry is secondary, visually de-emphasized.

Why

GSC data directly improves AI pre-selection accuracy in Steps 2 and 3. A user who connects GSC gives the system real search data — which pages get traffic, which topics are covered, which competitors appear. Manual URL gives much less. The hierarchy in the UI is a product quality decision, not just a UX convenience.

Risk

GSC requires a Google account with sufficient permissions. Users without access or without an established site are pushed toward a worse experience from the first screen.

4

Pages and Prompts combined in one step

Alternatives considered

  • Pages in Step 2, Prompts in Step 3 as two separate screens

What was chosen

Pages and Prompts combined in one step, side-by-side. Both AI pre-selected. One confirm button to continue — effective for the Fast Pass approach.

Why

Both depend on brand setup being complete but neither depends on the other. Keeping them separate could build a false impression that Pages must be confirmed before Prompts can be configured — a dependency that doesn't exist. Combining them removes that ambiguity, reduces the step count, and keeps the Fast Pass fast.

Risk

The combined view is more information-dense. Users who want to review everything carefully may feel rushed. The Manage panel (See all →) exists as the escape hatch, but it requires an extra click.

Next steps

1

Run usability tests with existing Surfer users

Current users already understand what the platform does and what value it delivers. They can quickly spot if something is misleading, if a label doesn't match their mental model, or if a default wouldn't work for their real use case. This is a different signal than unrelated users — less about comprehension, more about fit with existing expectations.

2

Run usability tests with unrelated users

Recruit a small number of people who have no prior context on Surfer or this project. The goal is to test whether the flow makes sense to a complete stranger — not someone who already understands the product concept. Watch where they hesitate, what they ignore, and whether they trust the AI pre-selection without questioning it.

3

Focus testing on the page and prompt selection step

This is the highest-uncertainty part of the flow. Observe whether users accept the AI defaults, open the Manage panel, or feel stuck. If most users just confirm without reviewing — is that confidence or confusion? That distinction matters for whether the Fast Pass approach is actually working.

4

Ship and collect baseline data

Once usability issues are addressed, ship the flow to real users and start measuring. The hypotheses in the Metrics page are only useful once there is real data to compare against. Everything before that is still a guess.

Ideas & future directions

Once baseline data is in from real users, these are the most promising directions. See the Metrics page for the full measurement framework behind these.

  1. 1
    Validate the page impact scoreCorrelate Step 2 pre-selections with actual ChatGPT citation data. If high-scored pages appear in citations more often, the ranking model is working and can be promoted as a feature.
  2. 2
    Instrument field edits in brand reviewTrack which AI-generated values get replaced and with what. Every edit is a training signal. This is how you improve brand extraction without a labelling team.
  3. 3
    Correlate onboarding method with 30-day retentionDo GSC users retain better? Do users who edit more prompts churn faster or slower? The onboarding data becomes a leading indicator for long-term product health.
  4. 4
    Segment by company typeOnce brand classification is reliable, split all metrics by segment. Same UI, different benchmarks. A blog and a SaaS company need different pre-selection logic.
  5. 5
    Activation as a primary metricCompletion rate tells you if users finished the flow. Activation tells you if the product kept its promise — did the user return to see their first AI visibility report within 48h? That's the real proof the onboarding worked.
  6. 6
    Quality / confidence as a secondary metricActivation measures return. It doesn't tell you if users set up tracking on the right things. A post-onboarding confidence score (even a simple 1-question prompt) would be the most valuable leading indicator for data quality.
  7. 7
    Measure friction at the page and prompt selection stepThis is the step with the least confidence. If a user doesn't trust the AI pre-selection, they face 50 pages and 25 prompts to review manually — which may cause drop-off or, worse, a blind confirm that leads to poor data quality. Time-on-step, manual edits, and panel open rate are the signals to watch here first.