AI Transparency Notice

GoTeam uses AI in two places: assessing applications in the employer panel, and in the CV Analysis report bought through goteamx.app. The flow is the same in both, so this notice covers them together.

The job of the AI is to gather evidence. It reads what the CV says, answers fixed questions and shows the sentence it relied on. The arithmetic that produces the score runs in code, not in the AI.

A score is not an opinion converted into a number. It is the share of a fixed evidence list that has been met. Asking the model "how many points would you give this candidate" would be asking it for a holistic judgement. That judgement changes from run to run on the same CV and there is no way to show what it rests on.

The assessment is AI assisted. The score is produced against a fixed, evidence based rubric, and the final decision always belongs to a person. If you believe that an exclusively automated analysis has produced a result to your detriment, you may object under Article 11 of Law No. 6698 (KVKK, the Turkish Data Protection Law).

In the panel the AI can suggest the next step. A suggestion is not applied on its own: the status of an application changes only when it is explicitly approved in the panel.

For a candidate who is seen not to meet a mandatory criterion, no interview is suggested, but no rejection is suggested either. The record is flagged separately and left to human review. The same route is followed where a criterion could not be verified from the CV, because a wrong rejection is more expensive than keeping an unnecessary candidate on the list.

The system does not learn from past hiring outcomes. Who was invited or hired never enters scoring as an input. A model that learns from past decisions reproduces the discrimination in those decisions.

The analysis is carried out through a model interface offered by our AI provider established outside Türkiye. What is sent is the text extracted from the CV and the criteria of the posting.

We give no undertaking on this page about whether the data is used in model training on the provider's side. We only write what the data is and where it is sent, because that is what we can verify from our own code. You can learn the identity of the provider through the channel described on the Data Subject Request page.

Which model ran is written into the analysis record. In the employer panel the model choice can be changed from the account settings; on the CV Analysis side the model is fixed by the product.

The full list of the providers that touch your data is published in the sub-processor list, and that list is the first place updated when it changes.

CV files are hosted by our cloud file storage provider established outside Türkiye and are sent to our AI provider established outside Türkiye for analysis. These two providers are the places where the data is processed outside Türkiye. The transfer falls within Article 9 of Law No. 6698 (KVKK, the Turkish Data Protection Law) and is necessary for the service to be provided; for the analysis service your explicit consent is obtained in addition. You can learn the identity of the providers through the channel described on the Data Subject Request page. If you do not give consent, no analysis can be carried out; on its own this leads to no other consequence.

Which personal data is processed for which purpose and on which legal ground, together with the retention periods, is set out in the KVKK Privacy Notice. For where the data is held and how it is protected, the Privacy Policy governs.

The checklist is applied to everyone in the same way and no item on the list names an occupation. That does not mean candidates from every occupation will receive the same score, and we make no such claim.

On 30 July 2026 real CVs from the same source, at the same seniority and from four different fields were run through the same pipeline. The language of the report came from each candidate's own occupation: the nurse got nursing terms and the criminologist got criminology terms, with no software template imposed.

In the same measurement the overall scores came out as 51, 51, 41 and 40. Almost all of the difference comes from the technical dimension. The cause is not the occupation itself but whether the CV contains a list of tools: a marketer who writes down the tools they use earns the same items as a software developer.

Today's behaviour is this: technical items that ask for a tool list, a project link and an online work sample stay in the denominator in occupations where that habit does not exist, and pull the score down when they cannot be earned. The limits of the measurement are equally clear: 4 CVs, all of them in English and at student level. It has not been repeated with Turkish or senior CVs.

The items below are not guesses; they are limits read from the code or seen in measurement.

• For a candidate application only PDF is accepted. A file in any other format does not enter the analysis.
• A scanned PDF with no text layer cannot be read. The file is rejected and no analysis credit is consumed.
• A document that is not a CV (an invoice, a thesis, a contract) is rejected at the classification step before the analysis. The classification is also done by a model and it can be wrong.
• If a CV is out of date, no calculation can see that. A candidate who writes "present" for a job they have left is credited with a longer period than they had. Checking that is left to the reference call.
• Instructions hidden in a CV, that is text addressed to the system rather than to the assessor, are scanned for. Detection does not stop the analysis: the record is flagged and goes to human review.
• Keyword stuffing in a CV could not be blocked completely in measurement. Document level stuffing detection works, but medium scale stuffing can be missed.
• The model can decide an item incorrectly. The requirement to quote word for word reduces that but does not remove it.
• In CV Analysis, a PDF of at most 15 pages is processed by default.
• Extraction and the report are produced in Turkish. An English CV is read, but the output is still Turkish.

On the application screen of the company you applied to, you can see what was read from your CV: experience titles, date ranges and technologies. The score, the band, the ranking and the gate decision are not shown to the candidate. The reason is to avoid creating expectations before a decision and to avoid teaching people how to game the score.

If what was read is incorrect, you can report a correction from the same screen. Your report does not delete the analysis and does not change the score; it moves the record to human review and writes your note into the audit trail. Allowing a candidate's own statement to move the score would make it possible to tune the system in one's own favour.

You may send your requests under Article 11 of Law No. 6698 (KVKK, the Turkish Data Protection Law) to [email protected]. Your request is concluded within 30 days at the latest. How the request is made and which information it must carry is set out on the Data Subject Request page.

Which model, which scoring version and which criterion list an analysis was produced with is on record. When someone asks what a decision rests on, the answer is given from that record. You can send your questions about this notice to [email protected] or to 0551 406 11 90.