# Testing AI Content Refreshes: Experiment Plan

Canonical page: https://2bcome.ai/experiments/content-refresh/
Content updated: 2026-09-15

EXPERIMENT 003 Planned test

# Can a focused refresh beat another new draft?

A proposed comparison of updating an existing page against creating more content without first resolving a known gap.

By [2bcome](https://2bcome.ai/about/) Written September 15, 2026 Updated September 15, 2026

Planned experiment. No results are available. This page describes the proposed test, not a completed case study.

## Status

Planned. No pages have been selected, changed, or measured for this test. There is no reported uplift.

## The question

Can the agent identify a specific weakness in an existing page and improve it with a focused revision, rather than defaulting to writing something new?

We want to test the quality of the diagnosis and the usefulness of the proposed change before making claims about performance.

## The proposed method

Choose a small group of pages with a defined purpose. Record their current content, page revisions, and relevant performance data.

Ask the SEO role to identify the gap. Ask the Content role to prepare a specific update using checked source material. The proposal must explain why this change is preferable to another new page.

Keep factual claims, examples, headings, and internal links reviewable. Do not rewrite unrelated sections merely to make the diff larger.

## A controlled release

Publish only approved changes. Preserve the original revision and log other site changes during the observation period.

A reference group may help the later review when a meaningful comparison is possible. It should not be presented as a perfect experiment merely because unchanged pages exist.

## What to evaluate

First, review whether the agent diagnosed a real problem, whether the draft addressed it, and how much repair was needed.

Second, evaluate the agreed performance measures after the observation window. Include missing data, alternative explanations, and inconclusive findings.

## What will be published

The original problem, the proposed revision, the approval record, the test conditions, and the observed result. A negative outcome should still produce an update to the workflow.

[Explore content optimization](https://2bcome.ai/solutions/optimize-existing-content/) or meet the [Content Agent](https://2bcome.ai/agents/content-agent/).

CONNECTED WORK

## The next useful connection.

WORKFLOW

### [AI Content Refresh and Optimization Workflow](https://2bcome.ai/solutions/optimize-existing-content/)

Explore a focused content refresh workflow: find outdated information, improve the existing page, review the diff, and measure against a recorded baseline.

[Explore](https://2bcome.ai/solutions/optimize-existing-content/)

SPECIALIST

### [AI Content Agent for Research and Drafting](https://2bcome.ai/agents/content-agent/)

Meet the planned AI Content Agent for source-backed briefs, useful drafts, and content refreshes. Review its inputs, outputs, and publishing boundaries.

[Explore](https://2bcome.ai/agents/content-agent/)

FIELD NOTE

### [How to Measure AI Agent Outcomes](https://2bcome.ai/blog/measuring-what-matters/)

How we intend to evaluate agent work without mistaking a busy log for a useful result.

[Explore](https://2bcome.ai/blog/measuring-what-matters/)