Wikivoyage:AI guidelines Voyage Tips and guide

This page is currently a draft. More information pertaining to this may be available on the talk page. It is not an official policy or guideline that reflects community consensus.

This is strictly a draft policy and this has not gained a consensus.

Artificial Intelligence (AI) programs can sometimes be quite helpful to Wikivoyage contributors, but you need to be very careful if you use one. There are several reasons that AI output should never be posted to Wikivoyage without human review:

• Many AI programs sometimes invent completely bogus "facts"; computer scientists refer to the phenomenon as "AI hallucinations."
• The commonest type of AI programs that generate text rely on Large Language Models (LLMs). These require a great deal of input, and some grab input off the web which may be under copyright. Critics call LLMs "plagiarism machines" and several copyright holders are suing various AI companies; as of early 2026 most of those cases are still before the courts, so the legal situation is not at all clear.

Note that the mediawiki software used here assigns copyright on all contributions to the contributor, so in theory a contributor using AI might be sued for claiming copyright on plagiarized material.

• Finally, the site has its own style, standard structures for articles and listings, and a set of policies on what belongs where. An AI program will not know any of this, so it becomes the user's responsibility.

If you use AI help in writing an article, you should at least check that there are no hallucinations and copy edit any generated text to suit Wikivoyage's style. Make sure every listing has an url that confirms the added information, and check the existence of any transport service you add. Using AI to find information is OK, but confirm it yourself; the prose is often better written by yourself, for style and to avoid potential copyright and plagiarism issues.

• Content — All reader‑facing material on Wikivoyage, including destination guides, listings, itineraries, transport details, safety notes, and images. Because this is practical travel information, accuracy and real‑world verification are essential.

• Discussions — All community spaces where editors collaborate, such as talk pages, policy discussions, and noticeboards. These must remain authentically human‑driven and free from automated influence.

• Artificial intelligence (AI) — Any tool or system that generates, rewrites, or meaningfully alters text without direct human authorship, including large language models and generative text tools.

• AI‑assisted content — Any content where AI contributes to the substance or meaning of the text. This requires disclosure and human review before being treated as stable.

• Mechanical edits — Low‑risk changes such as translation, grammar correction, spelling fixes, or minor language refinement that do not introduce new information. These do not require disclosure.

• Human review — A human editor verifying that AI‑assisted content is accurate, up‑to‑date, travel‑relevant, and compliant with Wikivoyage policies. Human review is required for all AI‑assisted content except mechanical edits.

• Travel‑relevant information — Listings, prices, opening hours, transport routes, safety advice, and other practical details that travellers rely on. AI may not generate or modify this without human verification.

• Disclosure — Clearly indicating AI involvement in an edit summary or talk page note whenever AI contributes to the substance of content or discussions.

Verification challenge — Wikivoyage has a limited number of active editors, which makes it difficult to reliably verify AI‑assisted content, especially when it involves real‑world travel details such as listings, prices, transport schedules, safety information, and opening hours. Because inaccurate travel information can mislead readers, AI use must be carefully restricted and clearly disclosed.

Topic limits — AI may be used only for low‑risk, background‑oriented sections where factual precision is less critical. Acceptable areas include history, culture, climate, geography, and general descriptions of a destination. AI use is prohibited for all practical travel information, including listings, prices, transport options, safety advice, visa rules, medical information, and anything requiring up‑to‑date, on‑the‑ground accuracy. These sections must be written or verified by humans with real knowledge or reliable sources.

Staged review — Any AI‑assisted content added to Wikivoyage must be clearly disclosed and marked for later verification by experienced editors. Until reviewed, such content should not be treated as stable or authoritative. Staged review ensures that AI‑generated text does not silently become part of the guide without human review, and it allows patrollers to prioritise which edits need attention.

Accuracy requirement — AI must not invent or guess travel information. This includes fictional hotels, restaurants, attractions, transport routes, prices, timetables, or safety warnings. All travel‑relevant details must be based on verifiable, real‑world facts, and editors are responsible for confirming accuracy before publishing.

No mass creation — AI tools may not be used to mass‑generate destination pages, itineraries, listings, or large volumes of new content. Wikivoyage prioritises quality, reliability, and local knowledge over rapid expansion. Any proposal for large‑scale AI‑assisted content creation must be discussed and approved by the community beforehand.

Mechanical edits allowed — AI use for purely mechanical tasks—such as translation, grammar correction, spelling fixes, and minor language refinement—is permitted without disclosure, provided the meaning is not changed and no new information is introduced.

Human responsibility — Editors remain fully responsible for all AI‑assisted content they publish. If AI introduces errors, outdated information, or misleading claims, the human editor who added the content is accountable for correcting it.

No AI images — AI‑generated images are not permitted on Wikivoyage. Only real photographs and media that accurately depict destinations, landmarks, and travel‑relevant subjects may be uploaded, ensuring readers receive authentic visual information.

• AI‑assisted comments — AI may help draft talk‑page comments, but editors must disclose AI involvement whenever it meaningfully shapes the substance, reasoning, or structure of what they post. This ensures transparency and prevents confusion about whether a comment reflects genuine human judgment.

• 'Human responsibility — Editors must fully understand, endorse, and be able to explain any AI‑assisted comment they publish. AI cannot be used to generate arguments the editor cannot defend or clarify, and the human editor remains accountable for tone, accuracy, and appropriateness.

• No consensus manipulation — AI must not be used to generate multiple comments, simulate support, or influence discussions through volume. Automated participation, rapid‑fire replies, or attempts to create the appearance of consensus are prohibited.

• No AI‑only arguments — AI‑generated reasoning cannot be posted without meaningful human involvement. Editors may not copy‑paste AI text wholesale or rely on AI to “speak for them”; discussions must remain authentically human‑driven.

• Mechanical edits allowed — AI use for translation, grammar correction, spelling fixes, and minor language refinement in discussion posts does not require disclosure, provided the meaning is unchanged and no new claims are introduced.

• Moderation assistance — AI may help identify vandalism, spam, or disruptive behaviour, but it must not take autonomous actions such as issuing warnings, closing discussions, or enforcing sanctions. All moderation decisions must be made by humans.

• No AI in formal closures — AI must not be used to draft or influence formal discussion closures, consensus summaries, or administrative decisions. These require human neutrality, judgment, and understanding of community norms.

• Local consensus priority — Wikivoyage’s established norms, policies, and community expectations override any AI‑generated suggestions or automated recommendations. AI cannot be used to reinterpret policy or assert authority in discussions.

Disclosure requirement — Editors must disclose AI involvement whenever AI contributes to the substance, structure, or meaning of any content or discussion post. Transparency allows patrollers and other editors to identify which edits require closer review.

Editors should name the AI tool, model, or company used (e.g., “GPT‑4”, “Claude 3”, “Gemini”, “DeepL Write”). This helps reviewers understand the likely behaviour and limitations of the system. When AI meaningfully contributes to content, editors should briefly summarise the prompt used. Full prompts are not required; a short description is sufficient. If a prompt contains private or sensitive information, editors should summarise it without revealing details.

Tool and prompt may be combined in one edit summary, for example,: [AI‑generated: Claude 3 | prompt: draft overview paragraph] This should be placed at the very start and should not be marked as a minor edit.

Once again no disclosure is required for mechanical edits such as grammar fixes, spelling corrections, translation, or minor language refinement that do not introduce new information.

Repeated failure to disclose AI use when required may be treated as disruptive editing, especially when it results in unverifiable or inaccurate travel information.

• First warning — A clear talk‑page warning explaining the disclosure requirement and asking the editor to correct future summaries. This assumes good faith and gives them a chance to adjust.

• Second warning — A stronger warning noting that repeated non‑disclosure disrupts patrolling and risks introducing unverifiable travel information. The editor is reminded that further issues may lead to editing restrictions.

• Editing restriction — If the behaviour continues, the editor may be restricted from adding AI‑assisted content entirely. They may still edit normally, but cannot use AI tools until trust is re‑established.

• Short block — For persistent or deliberate non‑disclosure, a short block (e.g., 24–48 hours) may be applied to stop the disruption and protect the accuracy of travel information.

• Escalation for abuse — If the editor repeatedly adds unverifiable or fabricated AI‑generated travel information, longer blocks or topic bans (e.g., “no editing destination pages”) may be considered.