AI for Journalists: Research and Transcription Workflows

AI for journalists is most useful in three areas: accelerating research on unfamiliar topics, transcribing interviews and press conferences, and structuring stories from messy notes. It should not write your stories, replace source verification, or generate quotes. Here is an honest task-by-task assessment for Kenyan newsrooms and freelance reporters.

What Journalism Tasks Look Like in Kenya

Kenyan journalism operates under specific pressures. Newsrooms have shrunk over the past decade while the demand for content has grown. A reporter at a Nairobi-based outlet may file three to four stories per day across digital, print, and broadcast. Freelancers juggle multiple outlets, each with different editorial standards and pay rates.

Time is the scarcest resource. A county government reporter covering Nairobi City County cannot spend three hours researching background for every story when three more stories are waiting. A business journalist covering the NSE needs to turn around earnings analysis quickly. A features writer has more time but faces the challenge of organizing hours of interview material into a coherent narrative.

AI fits into these workflows as a time-compression tool, not a content-generation tool. The distinction matters for reasons we will address throughout this guide.

Task: Background Research

What AI Does Well

When you are assigned a story on a topic outside your regular beat (a new agricultural policy, a legal dispute involving maritime law, a medical technology), AI provides a quick primer. Ask the AI to summarize the key concepts, identify the main stakeholders, and outline the relevant Kenyan regulatory framework.

"Summarize Kenya's Building Maintenance and Management Bill, 2024. Who introduced it? What does it propose? Who supports and opposes it? What are the main provisions?"

The AI gives you a starting framework in minutes rather than the hour of reading it might take to build the same understanding from scratch.

Where It Fails

AI research summaries are not primary sources. They are synthesized from training data that may be outdated or incomplete. A summary of a Kenyan bill may reflect an earlier draft, not the version currently before Parliament. The names of proponents or opponents may be inaccurate.

Every AI-produced research summary is a starting hypothesis. Verify key claims against official sources: the Kenya Gazette, Hansard records, official government websites, and direct contact with relevant offices.

Practical Workflow

1. Use AI to generate a background briefing on the topic.
2. Identify the three to five key claims or facts in the briefing.
3. Verify each one against a primary source.
4. Note what the AI got wrong or missed. This often points to the more interesting angle for your story.
5. Use the verified background to prepare better interview questions.

Task: Interview Transcription

What AI Does Well

Transcription is one of the highest-value AI applications for journalists. A 45-minute interview takes 2 to 3 hours to transcribe manually. AI transcription tools (Otter.ai, Google Recorder, Whisper, Trint) produce a draft transcript in minutes.

For English-language interviews, accuracy is typically most to nearly all. For Kiswahili, accuracy varies by tool (Whisper handles Kiswahili reasonably well). For interviews that mix English and Kiswahili (extremely common in Kenyan journalism), accuracy drops and you need to review more carefully.

Where It Fails

AI transcription struggles with Kenyan accents when speakers talk quickly, with Sheng, and with audio recorded in noisy environments (a market, a construction site, a matatu). Background noise from Nairobi traffic can make sections unintelligible to the AI even when you understood them clearly during the interview.

It also does not identify speakers reliably. In a multi-person interview or press conference, you need to manually attribute quotes to the correct speakers.

Practical Workflow

1. Record interviews with the best audio quality possible. A lapel mic or a phone placed close to the speaker significantly improves AI transcription accuracy.
2. Upload audio to your transcription tool immediately after the interview.
3. Review the transcript against the audio. Correct errors, especially in names, numbers, and Kiswahili or Sheng passages.
4. Attribute quotes to the correct speakers.
5. Highlight the key quotes and statements you plan to use in your story.

Task: Story Structuring

What AI Does Well

After a day of reporting, you have: interview transcripts, background research notes, press release text, and your own observations. Organizing this into a coherent story structure is a skill, but AI can speed up the process.

Feed your notes (anonymized if needed) to the AI and ask: "Based on these notes, suggest a story structure. What should the lead be? What are the strongest quotes? Where should the background context go?"

The AI's suggestion is a starting point, not a finished structure. It often identifies a lead angle you had not considered, or suggests organizing the story around a chronological narrative when you were planning a thematic structure (or vice versa). These alternative perspectives are valuable even when you do not adopt them.

Where It Fails

AI does not understand news judgment the way an experienced editor does. It cannot tell you that a particular quote will resonate with your audience because it captures a sentiment that has been building in public discourse. It cannot sense that a story's real significance is not in the facts presented but in what they imply about a larger trend. That editorial judgment is yours.

Practical Workflow

1. Compile your reporting materials as text (transcripts, notes, press releases).
2. Ask the AI for a structural suggestion.
3. Evaluate the suggestion against your own journalistic instincts.
4. Write the story yourself, using the structure as a guide where helpful.
5. The final story is entirely your work in language, judgment, and responsibility.

Task: Headline and Summary Generation

Headlines and social media summaries for published stories are a minor but time-consuming task, especially for digital journalists who need multiple versions (website headline, social media post, newsletter blurb, push notification).

AI generates these quickly. Feed the finished story and ask for five headline options and a 50-word summary. Pick the best one, adjust it for your publication's style, and publish.

This is a low-risk AI application because headlines are checked by editors and the source material (your own published story) is accurate.

What AI Must Never Do in Journalism

AI must not generate quotes. Fabricating a quote and attributing it to a real person is a firing offence in journalism. AI tools can generate plausible-sounding quotes that never happened. Never use AI-generated text as a quote from a source.

AI must not replace source verification. If AI tells you a fact, that is not a source. You need a verifiable primary source for every factual claim in your story. AI is a research tool, not a reporting tool.

AI must not write your story. Publishing AI-generated text under your byline without substantial original reporting is a breach of journalistic ethics. Your byline means you stand behind the work. You cannot stand behind work you did not do.

AI must not compromise source confidentiality. Do not enter confidential source information, leaked documents, or sensitive investigation details into public AI tools. The confidentiality of sources is a foundational journalistic principle. Entering this material into an AI tool that stores or processes conversations risks exposure.

Kenyan Newsroom Realities

Many Kenyan newsrooms cannot afford paid AI tools. The free tiers of ChatGPT, Claude, and Google Gemini are sufficient for research assistance and story structuring. Free transcription tools (Google Recorder, the free tier of Otter.ai) handle individual interviews, though they have time or usage limits.

For newsrooms with budgets, a shared Otter.ai or Trint subscription for the reporting team provides unlimited transcription, which is the single highest-impact AI investment for a newsroom.

Mobile phone access is often the primary tool for Kenyan journalists in the field. AI tools that work well on mobile browsers (ChatGPT, Claude) are more practical than desktop-only applications.

For journalists who want to build structured AI skills applicable to research and productivity workflows, the free welcome module of our AI and Automation course covers the fundamentals in a practical, tool-agnostic way.

FAQ

Does using AI for research make me a lazy journalist?

No. Using AI for background research is similar to using a search engine, an encyclopedia, or a research database. The research is a starting point, not the final product. Your reporting (interviews, source verification, on-the-ground observation) is what makes the story journalism rather than a summary. AI makes the preparation faster, which gives you more time for the reporting that matters.

Can I use AI transcription for off-the-record conversations?

Exercise extreme caution. AI transcription tools process audio on remote servers. If a conversation is off the record, think carefully about whether uploading that audio to a third-party service is consistent with your source's expectation of confidentiality. For highly sensitive conversations, manual transcription (or notes taken during the conversation) is safer.

How do I disclose AI use to my editor?

Be straightforward. "I used AI transcription for the interview" or "I used AI for initial background research, then verified against primary sources." Most editors care about the quality and accuracy of the final product. Transparency about your process builds editorial trust rather than undermining it.