Case Studies

Case Studies: Before & After

This page publishes measured results only: before/after scans taken with the same engine set, the same measurement depth, and the same brand definition. No figure that fails these conditions appears here.

Case 1 · Translation & localisation · B2B

Alafranga Language Solutions

turkishtranslationagency.com · 29 April → 26 June 2026 · 58 days

An Istanbul-born, London-based technical translation agency; ISO 17100 certified, operating since 2002. In April 2026 it was close to invisible in AI engines: mentioned in roughly 1 in 12 relevant queries, and never in a ranked position when it was.

Metric29 Apr26 JunChangeNoise margin
Mention rate (2 engines)%8,06%27,42+19,4 pp
Gemini%3,2%38,7+35,5 pp±4,0 / ±9,7
Claude%12,9%16,1+3,2 pp±6,8 / ±7,4
Average positionnot listed2,75entered ranking

93 queries per engine, one behaviour cohort. Noise margin: an engine's rate can drift within this ± range by chance; a change smaller than it is not treated as significant.

What the numbers say

  • The Gemini gain is far above the noise margin: +35.5 points against a widest margin of ±9.7. This is a real change.
  • The +3.2 points on Claude sit inside the noise margin. We do not count it as a result.
  • In April the brand held no ranked position in any answer; by June it averaged position 2.75. Beyond being mentioned, it entered the set of recommended options.

What changed on the site in this period

Three pieces of work landed on the site between the two scans. We cannot separate how much each contributed; a single scan pair cannot tell.

  • An llms.txt file for AI systems (19 June, seven days before the second scan): company definition, services, certifications and a citation format.
  • Structured data: Organization, Person, BreadcrumbList, ItemList and credential nodes (EducationalOccupationalCredential).
  • A deeper home page: services, sectors and process described on one page, roughly 3,500 words.

Limits

  • Two engines were measured: Claude and Gemini. ChatGPT, Perplexity and Google AI Overview are not in this pair; their state is unknown, not zero.
  • 93 queries per engine is a small sample. That is why the noise margins are wide, and they are printed in the table.
  • We do not attribute the Gemini jump to llms.txt. Google states it does not use llms.txt, and Gemini runs on Google infrastructure.
  • Share of voice is not compared: the competitor set differed between the two scans, so the denominator changed.
  • The platform's overall score rose from 2 to 30. We do not headline it — a composite score can rise while mention rate falls. For visibility we read only the per-engine breakdown.

Scan ids: 9fdab931-c645-4447-a323-1f7d6ed630d1 · 93fb91bb-b61f-4745-afc0-a037a6d003fc

Brand name and figures published with the client's permission.

Our own measurement series — unflattering numbers included — is also public: measurement series · methodology

Measure Your Site's AI Visibility

Start with our free GEO Readiness Index tool and see where you currently stand in AI search engines.

Start Free Analysis
WhatsApp Support