How It Works

The eight-step closed loop

Monitoring tools stop at step two. The last three — experiment, learn, optimise — are only possible once the change has actually been published, which is why they are the ones nobody else does.

Start Free TrialSee Pricing
  1. 1

    Observe

    Sample, do not spot-check.

    Each tracked prompt runs repeatedly per cycle, not once, because a single roll of the dice is noise rather than measurement. Every run is stored as its own event, and the citation rate comes back with a confidence interval rather than a bare percentage.

  2. 2

    Diagnose

    Find the page that is losing, and to whom.

    A prompt you are losing maps to a page that should be winning it. SearchForge scores every page on the gap between competitor and own citations, structure deficits, missing schema and staleness, then picks the ones worth working on.

  3. 3

    Generate

    Write the actual fix.

    Page copy, FAQ blocks, JSON-LD schema and answer snippets, grounded in retrieved context from the pages currently beating you. Output that cannot cite its grounding does not leave this step.

  4. 4

    Approve

    A human decides what ships.

    Quality gates run first — citations, answer snippets, FAQ count, meta length, grounding. Anything that fails is parked for review. Protected pages such as pricing and checkout always require approval, whatever the autopilot mode.

  5. 5

    Deploy

    Publish where the site actually lives.

    WordPress, GitHub, Wix, Squarespace, webhook or export — each through its own connector, each verified by re-fetching the live page afterwards to confirm the change really landed. One click reverts it.

  6. 6

    Experiment

    Record the change as an intervention.

    A publish is not just an edit, it is a hypothesis: this page change should influence these prompts. SearchForge writes that down — the page, the prompt set, the deploy time, and the baseline window preceding it.

  7. 7

    Learn

    Test the difference, do not eyeball it.

    After the post-intervention window, a two-proportion z-test compares the baseline citation rate against the current one. You get an effect size in percentage points and a p-value. When the sample is too small to conclude anything, it says so rather than guessing.

  8. 8

    Optimise

    Feed the result back in.

    A topic you have won cools down in the scoring, so the next cycle moves to the next gap instead of rewriting a page that is already working. This is the step that makes it a loop rather than a queue.

Already using Peec, Profound or Semrush?

Keep it. Those tools are good at step two. SearchForge is steps three through eight — import your citation gaps and we do the part that comes after the report.

Start Free Trial →