AI Policy

Last updated: August 27, 2026

Yazidis.info is an independent reference project about Yazidi history, religion, culture, heritage, communities, and contemporary experience.

Artificial intelligence can assist with research and editorial work, but it is not an authority, source, scholar, community representative, or substitute for human judgment.

This policy explains how Yazidis.info may use artificial intelligence and the safeguards we apply when AI contributes to our work.

Our Core Principle

Our approach can be summarized simply:

AI may assist the process. Evidence determines the content. Humans remain responsible for publication.

A statement does not become reliable because an AI system generated it.

Every substantive factual claim on Yazidis.info should ultimately be supported by an identifiable source such as an academic publication, primary document, archival collection, United Nations report, reputable NGO documentation, or appropriately documented Yazidi community source.

AI output itself is not treated as a source.

Why We Use AI Carefully

Generative AI systems can be useful for tasks such as organizing information, comparing documents, identifying research questions, or improving the structure of a draft.

They can also produce:

  • inaccurate statements;
  • invented citations;
  • false quotations;
  • oversimplified explanations;
  • outdated information;
  • mistranslations;
  • unsupported historical connections;
  • culturally inappropriate descriptions;
  • biased interpretations; and
  • confident answers where the underlying evidence is uncertain.

These risks are particularly significant for Yazidis.info because many subjects covered by this project involve religion, minority identity, genocide, persecution, oral traditions, contested history, and community perspectives that outsiders have frequently misunderstood.

NIST’s Generative AI Profile specifically treats generative-AI risks as something organizations should identify and actively manage rather than assuming generated material is trustworthy by default.

Human Editorial Responsibility

A human editor remains responsible for material published on Yazidis.info.

We do not delegate final editorial responsibility to an AI system.

AI may propose text, identify possible sources, organize notes, or highlight inconsistencies, but publication decisions should ultimately be made through human review.

This principle reflects a broader standard in responsible AI governance: UNESCO’s Recommendation on the Ethics of Artificial Intelligence emphasizes that AI systems should not displace ultimate human responsibility and accountability.

How AI May Be Used

Yazidis.info may use AI-assisted tools for limited editorial and technical purposes, including:

Research Discovery

AI may help identify:

  • possible books or papers to investigate;
  • alternative spellings of names and places;
  • terminology used in different scholarly traditions;
  • relevant organizations or archives;
  • questions that require further research; or
  • potential gaps in an article.

Any source suggested by AI should be independently verified before being relied upon.

Organizing Research

AI may assist with:

  • grouping research notes;
  • organizing timelines;
  • comparing terminology;
  • creating preliminary outlines;
  • identifying recurring themes;
  • summarizing material for internal review; or
  • highlighting apparent contradictions between sources.

The original source remains authoritative over an AI-generated summary.

Drafting Assistance

AI may help create or revise preliminary text, including:

  • article outlines;
  • introductory drafts;
  • transitions;
  • explanatory passages;
  • headings;
  • FAQs;
  • metadata;
  • descriptions; or
  • alternative wording.

AI-generated draft text is not automatically publishable.

Factual statements must still be reviewed against appropriate sources.

Copyediting

AI may assist with:

  • spelling;
  • grammar;
  • readability;
  • sentence structure;
  • consistency;
  • formatting; and
  • repetitive wording.

Editorial improvement must not alter the meaning of a quoted or sourced statement.

Translation and Transliteration Assistance

AI tools may sometimes assist with preliminary translation or transliteration.

However, AI-generated translations can be unreliable, particularly with:

  • Kurmanji Kurdish;
  • Arabic;
  • historical terminology;
  • religious vocabulary;
  • proper names;
  • regional expressions; and
  • oral religious material.

Sensitive or consequential translations should therefore be checked against authoritative translations, dictionaries, specialists, community sources, or native speakers where reasonably possible.

Machine-generated translation should never silently be presented as authoritative interpretation of Yazidi religious terminology.

AI Is Never a Citation

Yazidis.info does not treat ChatGPT, Gemini, Claude, Copilot, or another generative-AI system as evidence for a historical, religious, cultural, demographic, legal, or scientific claim.

For example, we should not publish:

According to an AI system, this Yazidi tradition originated in…

Instead, the underlying evidence must be identified.

Where a claim originates in an academic monograph, the monograph should be cited.

Where it originates in a UN investigation, the UN documentation should be cited.

Where it represents a Yazidi community understanding, an appropriate Yazidi source should be identified whenever possible.

Citations Must Be Verified

AI systems sometimes generate bibliographic references that look convincing but do not exist.

For this reason, an AI-generated citation should never be trusted simply because it contains:

  • an author;
  • a title;
  • a journal;
  • a publisher;
  • a DOI; or
  • a page number.

Before publication, important references should be checked to confirm that:

  1. the source actually exists;
  2. the author and title are correct;
  3. the cited material genuinely supports the statement;
  4. the publication details are accurate; and
  5. the source is appropriate for the claim being made.

Fabricated citations are unacceptable.

AI Must Not Invent Quotations

Direct quotations require special care.

AI systems may paraphrase material while presenting it as if it were a quotation or may generate wording that never appeared in the source.

Therefore, quotations should be checked against the original publication or a reliable reproduction before publication.

Quotation marks should not be placed around AI-generated reconstructions.

Special Rules for Yazidi Religion

AI systems have been trained on large quantities of internet material containing inaccurate or sensational descriptions of Yazidi religion.

Consequently, AI-generated explanations of Yazidi theology, Tawûsê Melek, Lalish, religious castes, sacred narratives, rituals, taboos, festivals, or oral traditions should be treated only as starting points for research.

They must not be accepted as authoritative merely because the generated explanation sounds confident.

When academic interpretations and Yazidi understandings differ, we aim to identify both rather than asking AI to decide which is correct.

AI must not be used to manufacture a supposed Yazidi belief for which no credible community or scholarly source can be identified.

Special Rules for Genocide and Survivor Testimony

Yazidis.info covers the genocide committed against the Yazidis and other traumatic subjects.

AI should be used with heightened caution when working with:

  • survivor testimony;
  • sexual violence;
  • captivity;
  • killings;
  • missing persons;
  • mass graves;
  • children;
  • personal identities;
  • medical information;
  • family details; or
  • other potentially sensitive information.

AI-generated text must never be allowed to invent details about a survivor’s experience.

Testimony should be represented according to the original source.

AI should not merge accounts from different survivors into a fictional composite unless an article explicitly describes such a methodological exercise and there is a compelling legitimate reason for doing so.

Respect for Community Perspectives

AI models may reproduce biases present in historical scholarship or on the public internet.

For a minority community such as the Yazidis, this can result in an outsider’s interpretation being presented as though it were the community’s own understanding.

Yazidis.info therefore tries to distinguish among:

  • what Yazidis say about themselves;
  • what particular Yazidi individuals or institutions say;
  • what scholars argue;
  • what historical outsiders claimed;
  • what governments or international organizations concluded; and
  • what remains uncertain.

AI should not collapse these categories into a single supposed “Yazidi view.”

This approach is consistent with UNESCO’s broader emphasis on fairness, non-discrimination, human dignity, transparency, and human oversight in the use of artificial intelligence.

Disagreement Must Not Be Artificially Resolved

If reliable sources disagree, an AI system should not be used to manufacture consensus.

Instead, an article should explain the disagreement where it is relevant.

For example:

Scholar A argues X. Scholar B proposes Y. Available evidence does not currently establish a universally accepted conclusion.

This is preferable to asking an AI system to choose whichever explanation appears most probable.

AI-Generated Images

Yazidis.info may occasionally use AI-assisted or AI-generated graphics for illustrative or design purposes.

Such images should not be presented as authentic historical photographs, archaeological evidence, documentary records, or representations of actual people or events when they are not.

AI should not be used to create fictional documentary images of:

  • genocide scenes;
  • identifiable survivors;
  • religious ceremonies represented as historical documentation;
  • archaeological objects represented as genuine finds; or
  • historical events in a manner likely to mislead readers.

Where an AI-generated image could reasonably be mistaken for documentary evidence, it should be clearly identified as generated or illustrative.

NIST identifies content provenance as one mechanism that can help audiences distinguish synthetic from authentic content.

Historical Photographs

AI restoration or enhancement of a historical photograph should be approached differently from ordinary design graphics.

Techniques such as:

  • colorization;
  • face reconstruction;
  • generative filling;
  • object removal;
  • background reconstruction; or
  • creation of missing image areas

can introduce details that were never present in the original historical record.

When such processing materially alters the evidence visible in an image, the modification should be disclosed.

The unaltered source image should be preferred when historical authenticity is important.

No AI-Generated Sources or Evidence

Yazidis.info will not intentionally manufacture:

  • archival documents;
  • eyewitness testimony;
  • scholarly citations;
  • historical photographs;
  • quotations;
  • oral traditions;
  • interviews;
  • demographic statistics;
  • religious texts; or
  • archaeological evidence

with AI and present those materials as authentic.

Synthetic material must not be disguised as primary-source evidence.

Personal Information and Privacy

Sensitive personal data should not be casually submitted to generative-AI systems.

Particular caution should be used with information involving survivors, children, missing persons, witnesses, religious minorities, or individuals who may face security risks.

UNESCO’s AI ethics framework identifies privacy and data protection as core considerations throughout the AI lifecycle.

Where possible, unnecessary identifying details should be removed before material is processed through third-party AI tools.

Automated Publishing

Yazidis.info does not consider mass production of automatically generated pages to be a substitute for research.

Automation may assist editorial workflows, but pages intended to function as reference material should add genuine informational value and receive appropriate review.

Google’s current Search guidance similarly states that generative AI can be useful for research and structuring content but warns that generating large numbers of pages without adding value may violate its policies on scaled content abuse.

Our standard is therefore not:

Was this written by a human or AI?

The more important questions are:

Is it accurate?

Is it useful?

Is it properly sourced?

Does it represent uncertainty honestly?

Has someone taken responsibility for publishing it?

AI Detection Tools

Yazidis.info does not treat automated “AI detectors” as definitive evidence that text was or was not produced using artificial intelligence.

Such tools may produce false positives and false negatives.

Editorial review should focus primarily on sourcing, accuracy, originality, transparency, and responsibility.

Disclosure

We do not necessarily label every instance in which AI assisted with routine tasks such as grammar correction, formatting, internal research organization, or metadata suggestions.

However, additional disclosure may be appropriate when AI plays a substantial role in material where knowing that fact would help readers evaluate the content.

This may include significant synthetic imagery, AI-generated reconstructions, or unusual forms of automated analysis.

Google’s guidance likewise notes that providing readers with information about how automated content was created can offer useful context.

Corrections Apply Regardless of How an Error Was Created

AI involvement does not excuse an error.

If AI-assisted content results in:

  • a factual mistake;
  • a fabricated citation;
  • a mistranslation;
  • misleading terminology;
  • inappropriate generalization;
  • incorrect attribution; or
  • misrepresentation of a Yazidi perspective,

we treat the issue as an editorial error and correct it according to our normal correction process.

Corrections from Yazidi readers receive priority in our editorial queue.

Our Editorial Standard

Before AI-assisted material is treated as reliable reference content, the relevant editor should ask:

  • What is the source for this claim?
  • Does the cited source actually say this?
  • Is the source appropriate and sufficiently authoritative?
  • Is this fact or interpretation?
  • Is scholarly disagreement being hidden?
  • Does this represent a community perspective or an outsider’s interpretation?
  • Could the wording reinforce an old misconception?
  • Has a translation changed the original meaning?
  • Is sensitive information being handled appropriately?
  • Would a reasonable reader be misled about how this material was created?

If those questions cannot be answered satisfactorily, the material should not be published as established fact.

A Tool, Not an Authority

Yazidis.info does not reject artificial intelligence simply because it is artificial.

Nor do we trust it simply because it is sophisticated.

AI can be useful for research support, organization, language work, and editorial assistance. Responsible use requires recognizing its limitations.

Our governing principle remains:

The source matters more than the generated sentence.

Human editors remain accountable.

Community representation requires care.

Uncertainty must remain visible.

And no AI system gets the final word on what Yazidis believe, what history proves, or what a source actually says.