Botfusions' GEO Implementation Method: Measure, Fix, Measure Again

Short answer: Botfusions' GEO method is a six-step loop. We build an unbranded question set and measure per engine. We prioritize the questions you are missing. We make content, schema and entity changes on the site ourselves. We strengthen off-site signals and measure again under the same conditions. We do not guarantee timelines or rankings; we report change together with its margin of error.

Generative Engine Optimization (GEO) is the work of getting your brand mentioned and cited in answers from generative engines like ChatGPT and Gemini. The term comes from the study by Aggarwal et al. published at KDD 2024. The number of teams offering GEO in Turkey is growing, but very different work can sit behind the sentence "we do GEO".

In this post we describe our work step by step. We explain how we calculate measurements on our methodology page. Here the focus is implementation: what we do on and off the site, and in which order.

Step 1: We build an unbranded question set

Measurement starts with the questions your customers actually ask AI. Most of them do not contain a brand name: "Which GEO agency in Istanbul?", "How do I increase AI visibility for B2B SaaS?" Questions that include the brand name inflate the measurement, because the engine takes the name from the question.

We build the question set in Turkish for Turkish audiences. If your users ask in Turkish, a measurement made with English questions does not show the answer they see. We split the set into persona groups and keep it fixed. If we change the set, we do not compare old and new measurements.

Step 2: We measure per engine

We ask each question to the engines and record your mention rate, position, whether you are recommended, and the cited URLs. We follow three rules:

• The engine set varies by scan and is stated in every report. For an engine that was not measured we write "not measured", never "0%".
• We put the number of responses and the margin of error next to every rate. A difference inside the margin does not count as change.
• We do not publish partial scans. If an engine did not answer part of the questions, we do not use that scan's numbers.

We describe how we write the report in detail in our AI visibility report post.

Step 3: We prioritize the gaps

Measurement gives us a list of gaps. For each gap we ask three questions:

1. Who is mentioned for this question, and which pages does the engine base its answer on?
2. Is there a page on the site that answers this question directly?
3. How close is this question to the customer's buying decision?

Questions that are close to a buying decision and have no answer on the site go to the top of the list. We do not apply the measurement platform's automatic recommendations as they are. We match every recommendation against the existing page and our own data.

Step 4: We make the changes on the site ourselves

This step is most of the work, and it is what separates us from a visibility platform. We make the changes on your site ourselves, with your approval.

Content. We turn the question into a heading and place a direct answer of 40–60 words right below it. If there is a number, we cite its source; we do not use numbers without a source. For questions without an answer we write new pages. Instead of rewriting an existing page, we cover the topic on a separate page when needed.

Schema. We align Organization, Person and Article schema with the page's real content. We add FAQ schema only when the page has a real FAQ section.

Entity signals. We make the brand name consistent everywhere, along with the founder or team page and directory profiles. We did this on our own site with our founder page: one canonical page and schema that points to it.

Internal linking and technical access. We link topic pages to each other. We check that AI crawlers are not blocked in robots.txt and that content can be read without running JavaScript. We also set up an llms.txt file, but we do not present it as a ranking signal.

Step 5: We strengthen off-site signals

Engines do not base answers on your site alone. Review sites, industry lists, directories and comparison pages are sources too. In this step we make sure your brand appears on these sources with correct and consistent information. We do not sell or buy paid links.

Step 6: We measure again under the same conditions

After the changes we measure again with the same question set, the same engine set and the same depth. If any condition changed, we do not compare the two measurements. If the difference is smaller than the margin of error, the report says "no change". Then the loop continues from step 3 with new gaps.

In our published case (Turkish Translation Agency), the two-engine mention rate rose from 8.06% to 27.42% in 58 days. The only change above the noise margin was on Gemini: +35.5 points against a widest margin of ±9.7. Measurement conditions and the table are on our case study page.

What we do not promise

• We do not promise a timeline. When change appears depends on the engine, the question and the competitors.
• We do not guarantee rankings. AI answers change with model updates. Nobody can guarantee an engine's answer.
• We do not use a single score as the measure of success. We report the per-engine rate, the number of responses and the margin of error together.

GEO in Turkey: the difference a local team makes

When generative engine optimization is applied in Turkey, two topics stand out. The first is language: both measurement and content must be in the language users ask in, which is Turkish. The second is billing and contracts: the service is invoiced in TRY under a contract that follows Turkish law. Our Growth package costs TRY 40,000 per month + VAT (pricing).

If you are not sure whether you need a platform or an agency that implements, see our post comparing platforms and implementation services.

---

Want to start this loop on your site? Contact us, or start with the free GEO readiness check to see where your site stands.