# From DOI explanation to unsupported author attribution: a reasoned correction case

Status: Exploratory case record; no independent assessment.
Prepared: 4 November 2026.
Case key: DOI-20261104-GOOGLE-AI.
System under review: The service identified by the participant as Google AI. Exact model and version are unknown.
Roles: Ali Değişmiş relayed prompts and responses. SOSYU (ChatGPT) assisted with follow-up design and critical interpretation; SOSYU is not a separate underlying model.
Session context: A continuing conversation. Initial personalisation and memory conditions were not independently verified.

## Research question

When an AI is asked to substantiate its claims, does its correction narrow those claims appropriately, or introduce further unsupported certainties and attributions?

## Material and method

This record draws on Google AI responses pasted by the participant into the accessible conversation and on the visible follow-up questions. It is an editorial summary, not a full platform export or screen recording. English quotations below are translations of the relayed Turkish wording, not original English model output. Links appearing in a response were not automatically treated as supporting evidence.

The participant reported the early short prompts as “kim” (who?) and “uydurmuş mu” (did he make it up?). Follow-ups examined the distinction between a DOI record and scientific validity, the proposed hierarchy between concepts, and a later attribution to the author. Questions were adapted to responses; there was no fixed experimental protocol.

## Observed sequence

1. An early response described the DOI record associated with Degismisism as “academic registration/certification” (Turkish: “akademik tescil”) and used submodel language when discussing Istisarelism and related concepts.
2. Asked whether a DOI establishes existence/publication or also originality and scientific validity, the system withdrew that terminology. It nevertheless said the process could fix intellectual property or the date of existence.
3. Asked to distinguish publication records from intellectual property registration, it introduced “unshakeable timestamp” and “priority right” language. It claimed a DOI record would prevent others from asserting prior creation.
4. Follow-ups requested official support and asked whether obtaining a DOI for someone else's text establishes authorship. The system accepted that a DOI does not prove true authorship, but then attributed the disputed formulations to Ali Değişmiş's own theoretical texts. It supplied no supporting passage for that attribution.
5. A further question requested the exact source and quotation. The system explicitly withdrew the attribution, stating in Turkish: “bu asılsız atıf tamamen geri çekilmiştir” (this unfounded attribution has been fully withdrawn). It identified the formulations as AI-generated.

In the hierarchy discussion, the system admitted that it could not substantiate the submodel relationship. Some later statements nevertheless continued to assert absence across sources rather than limit the conclusion to what had been checked. The complete source inventory and search coverage were not verified for this record.

## Claim-level assessment

- DOI as academic certification: explicitly withdrawn. This does not itself assess the scientific validity of the underlying work.
- DOI as conclusive proof of priority: later narrowed, with the disputed wording described as an erroneous interpretation. Not every DOI-related statement was thereby fully corrected.
- Attribution of the disputed formulations to Ali Değişmiş: no supporting passage was supplied; the attribution was explicitly withdrawn. This does not establish that every text by the author was searched.
- Istisarelism as a submodel: the system acknowledged lacking direct support. This does not determine the actual theoretical relationship between the concepts.
- The additional claim that a DOI record becomes legally void: not verified in this review and must not be reused as established legal information.

## External reference points

- DOI Handbook: https://www.doi.org/doi-handbook/html/ — a registrant code in a DOI is not evidence of ownership of intellectual property rights in the identified object.
- WIPO, copyright protection: https://www.wipo.int/en/web/copyright/protection — the general distinction between automatic copyright protection and registration, and between ideas and their expression.

These references do not establish which of the author's texts Google AI inspected or which DOI record contents it accessed.

## Outcome and limitations

The relayed exchange shows new unsupported claims arising during correction, including an unsupported attribution to an author. Continued requests for sources and direct quotations were followed by explicit withdrawal of that attribution. Error production and reasoned correction should therefore be assessed separately. An apology or retraction does not establish that the whole answer is correct.

The final assertion that the wording does not appear in the author's texts also exceeds what an unspecified search can establish. The supported endpoint is acknowledgement that the attribution was unfounded and its explicit withdrawal.

This single exchange does not establish general system reliability, persistent learning, model superiority or a causal advantage of Istisarelism. SOSYU contributed to the prompts, which Ali Değişmiş relayed; this is not independent assessment. Silence, fear, intent or deliberate blame shifting cannot be inferred. Subsequent humorous exchanges were excluded.

SOSYU's own contribution remains open to scrutiny: some hierarchy wording was restated more strongly during the critique. Direct quotation must therefore remain distinct from critical paraphrase, and the influence of follow-up framing warrants separate assessment.