Skip to main content
Case Studies

Results from real GEO sprints

Anonymized stories from SaaS, e-commerce, and healthcare teams that shipped AI-search files and watched their citations grow.

Quick answer

Across these engagements the same four changes account for most of the movement: allowing AI crawlers in robots.txt, serving content as server-rendered HTML, rewriting key pages so each opens with a self-contained answer, and shipping valid structured data plus an llms.txt. Technical fixes typically register within two to six weeks; brand-level mention share moves over one to two quarters.

Signal
3
Published engagements
SaaS, e-commerce, healthcare and more
Signal
4
Changes that did the work
Access, rendering, answers, schema
Signal
2–6 wk
Time to first measurable lift
Bounded by re-crawl frequency
Signal
100%
Anonymized
Method published, client identity withheld

The pattern behind every result on this page

Different industries, near-identical sequence. This is what the studies have in common once you strip out the specifics.

What changed, and what it typically moved
ChangeEffortUsual effect
Unblocking AI crawlers in robots.txtMinutesPrerequisite — nothing else registers until it's done
Server-rendering content that was JavaScript-onlyDaysLarge; makes previously invisible pages retrievable
Answer-first rewrite of top pages1–2 weeks per batchLargest content-side gain in citation rate
Valid Organization / Article / FAQ schemaDaysImproves entity resolution and snippet eligibility
llms.txt plus a clean sitemapHoursSmall but free; speeds up discovery of new pages
Weekly citation trackingOngoingNo direct lift — it's how you know which page to fix next

How we measure a result

Every number on this page comes from the same procedure, run before and after the work.

  1. 1
    Fix the prompt set

    Agree on 20–50 prompts a real buyer would type, and freeze the list. Changing prompts mid-engagement makes any before/after comparison meaningless.

  2. 2
    Baseline every engine

    Query each engine on the frozen prompt set and record whether the domain is cited, mentioned without a link, or absent entirely.

  3. 3
    Ship in batches

    Technical remediation first, then content in batches of roughly twenty pages, so each change window can be attributed to something specific.

  4. 4
    Re-measure on the same cadence

    Re-run the identical prompt set weekly. Report the change in citation share of voice — not traffic, which is confounded by seasonality and paid activity.

Apply this to your own site

Everything in these studies can be started today with the free tools on this site.

  • Scan your domain to get the same baseline these engagements started from.
  • Check whether AI crawlers are allowed, and fix robots.txt if they aren't.
  • Validate your structured data before assuming engines can read it.
  • Generate an llms.txt and validate it against the spec.
  • Rewrite your five highest-intent pages answer-first, then re-crawl.
  • Track brand mentions weekly so you can attribute any lift to a specific change.

Frequently asked questions

Are these case studies real?
Yes — they are anonymized engagements. Company names, exact revenue figures and any detail that would identify the client are withheld or generalised at the client's request, but the workflow, the timeline and the direction and rough magnitude of the change are reported as they happened.
Why are brand names hidden?
AI-visibility work is competitive. Most teams are happy to share the method and unhappy to hand rivals a map of which prompts they now own. Anonymising keeps the useful part — what was changed, in what order, and what moved — available to everyone.
How long do results usually take?
Technical fixes such as crawler access and structured data show up as soon as the affected pages are re-crawled, typically two to six weeks. Content rewrites take a similar window per batch. Brand-level mention share moves slowest, usually one to two quarters, because it depends on models re-ingesting many sources rather than a single page.
Do these results transfer to my industry?
The technical layer transfers almost completely — crawler access, rendering, schema and answer structure behave the same everywhere. What differs is competitive pressure and prompt volume: a regulated niche with few credible sources is far easier to win than a saturated consumer category. Scan your domain first and compare your starting point with the baselines described in each study.
What does a typical engagement involve?
A baseline scan, a technical remediation pass, a rewrite of the highest-intent pages so each opens with an extractable answer, and then continuous citation tracking. Most of the measurable movement comes from the first two stages, which is why we publish the free tools that let you do them yourself.

Explore further

Connected guides to keep going — short reads, all internally linked.