Our AI Policy

Last updated 15 October 2026

Since the first Artificial Intelligence (AI) chatbots were released, our reporters and editors have experimented with this new technology to see how it could enhance our work.

AI technology has improved dramatically over the past few years. We are increasingly using AI. This policy aims to guide our contributors on how to benefit from this technology while keeping human contributors central to GroundUp’s work and mitigating risks. The policy will be updated by the GroundUp editors as the technology develops.

“AI tools” in this policy refer to any generative AI technology, such as large language models (LLM), speech models, and diffusion (image generation). These include those used as a chatbot (e.g. ChatGPT, Claude, Meta AI), a built-in grammar or editing tool (e.g. Grammarly, Microsoft Word), or AI-powered search (e.g. Google Gemini, Google AI summary).

Guiding principles

Human beings remain the primary source of intelligence and creativity at GroundUp.
We prefer receiving pitches and articles that have been written by humans, rather than computers.
If AI has been used to draft or rewrite sentences, summarise sources, or present facts, this must be disclosed to the editors.
AI is not a source. Statements of fact must always be substantiated with the primary or original sources.
AI should not degrade the quality of GroundUp’s work or lead to the publication of false or misleading information.
Human journalists and their editors remain ethically and legally responsible for any mistakes that AI makes.

Using AI for research

AI tools can be powerful research tools. They can search through loads of information in a short amount of time. But they must not be trusted as a source. They can make things up or claim that information is contained in a source when it isn’t. AI tools can also draw incorrect conclusions and miss important facts. Extra caution is needed to avoid this.
Reporters who use AI for research should fact-check what they are told by checking the original source of the information: is the information the AI tool referred to actually there? Do you, as the reporter, fully understand it?
If a story requires reading large amounts of information – long documents or court judgments, for example – caution must be taken if using AI for summarising. AI summaries of long documents can omit important information, and we recommend that reporters still read most if not all of the documents. If this is not the reporter’s strength, they are encouraged to ask the editors for assistance.
Confidential material should not be uploaded to an AI tool unless the tool runs entirely on your own device, or GroundUp's editors have approved it. To be approved, a tool must not keep the material, must not use it to train its models, and must not give anyone outside GroundUp access to it (zero-data retention).
If at any stage a reporter is unsure about whether or not the information they have been given by the AI tool is correct, they can ask an editor for assistance.
When submitting a story, links to the original sources must be included in the submitted copy.

Writing

GroundUp prides itself on quality on-the-ground reporting. Humans are currently better at this than any AI tool. We require our reporters to have the skills to identify stories, pitch them and report on them without reliance on AI.
The primary creative force behind finding leads, identifying news angles and outlining story ideas should be a human being. AI tools could be used throughout this process: sharpening an angle, drafting interview questions, identifying who to contact or what records to request. But the reporter remains responsible for the judgment calls.
We prefer articles that have been written by humans, rather than AI. If AI is used in the writing process, we ask that reporters submit their notes or their draft from before the AI pass, so editors can compare.
The human authors and editors, not AI, will be held liable for any errors, misquotes, plagiarism or false citations.
Using AI tools to clean up written copy must be done with extreme caution, with specific, scoped prompting. AI writing sometimes introduces new facts, makes things up, or changes direct quotes.

Editing

GroundUp’s editors have developed an LLM tool that reviews articles at various stages of the editing process. This has been very helpful in catching typos and grammatical errors, finding internal inconsistencies or contradictions against previous reporting, fact checking the articles against publicly available information, and catching any holes in the story the editors may have missed.
The human editors are still the main drivers of the editing process and are responsible for any judgment calls made. The editors should take extra care to ensure the use of AI does not lead to unintentional changes in GroundUp’s tone and style.

Transcription

There are several very good AI transcription tools that create accurate transcriptions of interviews.
Extra care must be taken to ensure that direct quotes correspond with the actual audio.
If you are interviewing someone who wants to remain off-the-record or anonymous, then using an AI tool to transcribe their interview is unethical as the AI company may be given access to that information.

Graphics and Photos

GroundUp prefers photos and illustrations created by humans over images generated by AI. Captions under AI-generated images must indicate what AI model was used.
AI images may only be used as clearly illustrative graphics, never as realistic pictures of real people, places or events.
AI may not be used for generative alteration of news photographs.

Videos

The main concept/thesis of a video should preferably originate with a human, and so should the script. AI may be helpful to aid this process, but humans should lead it.
If AI tools have been used to generate graphic animations, text on screen must indicate the AI model that was used.
AI tools should not be used to narrate or present videos — we have sufficient skill and capacity to do this in-house.
AI tools should not be used to clone or synthesise a real person's voice, including AI dubbing of an interviewee into another language.

Programming

GroundUp has developed extensive IT intellectual property (albeit nearly all of it published under open source licences). Our systems are leading-edge by the standards of small newsrooms.
In 2026, we began using LLMs to assist with coding and configuration of our IT systems. These have been extremely useful. They have helped reduce bugs and increase productivity. But a significant amount of our software development is still done by hand. This is likely to remain the case for some time.
The principles we follow are (1) a human should ideally understand the architecture of public facing apps or programs we develop and (2) currently no LLM may be installed on or have update access to our main news server.

Choosing tools: ethical considerations

AI tools come with ethical costs. Data centres use large amounts of energy and water. Many models were trained on journalism and other creative work without permission or payment, sometimes including GroundUp's own. Some AI companies rely on poorly paid workers to label data. And control of the technology is concentrated in a few large companies.

We remain conscious of the ethical costs, but need to weigh them against accuracy, efficiency and affordability. An LLM with the least ethical cost is probably one hosted on our local machine, but running a model that is powerful enough locally is currently unaffordable for us.

No frontier AI company excels at transparency, privacy or environmental cost. GroundUp currently has paid subscriptions for Claude. Under this subscription, Anthropic, which makes Claude, does not use our data to train its models. We cannot claim Claude is the most ethical choice. Like other leading AI companies, Anthropic does not disclose its energy, water or carbon footprint. It also downloaded millions of pirated books and has been scanning and destroying rare books.

We will review our choice of tools at least once a year and move to less harmful alternatives when they become available and affordable.

Why we’re not blocking AI from our site

We are concerned about the failure of AI companies to seek permission before using data to train their models, usually without any payment, reward or recognition of the sources of that data.

Nevertheless, GroundUp has always had the policy of making our content as widely available as possible, at no charge. Moreover, given our size vis-a-vis the size of the large AI companies, we believe that excluding our content from AI models would simply reduce the visibility of our articles and GroundUp’s footprint. GroundUp’s articles are already providing users with valuable, reliable information about human rights in South Africa via mainstream chatbots that would otherwise not be available. We will review this policy regularly.