Effective: October 22, 2026 · Last updated: October 22, 2026
AI helps us work faster. A person decides what is true.
AI Visibility Journal (AVJ) studies AI systems, and we use them in our own work. Readers deserve to know exactly how. This policy applies to everyone who works on AVJ, including contributors and contractors.
Where we use AI
We use AI tools to:
- find and summarize background material, which a person then reads in the original;
- suggest outlines, headlines, summaries and metadata;
- write first drafts of descriptive or explanatory passages from a brief, which a person then rewrites and edits;
- extract every factual claim from a draft and check it against the cited sources, as a first pass that flags problems for a person to resolve;
- suggest line edits for grammar, consistency and style;
- review research designs and analysis plans, as one input to our own judgment;
- help write and debug analysis code, which a person reviews and runs;
- prepare and schedule posts on our official social media accounts that summarize articles we have already published (see below).
When AI-system outputs are the subject of our research (for example, which sources a chatbot cites), we collect them as data. They are never treated as a source of facts.
We use AI models from more than one company, and the model that checks a draft is from a different company than the model that helped write it.
AI at each stage
| Stage | What AI does | What a person does |
|---|---|---|
| Story selection | Scans announcements and studies and suggests items worth covering | The editor-in-chief opens the primary source and decides whether to cover it |
| Brief and research | Suggests outlines and summarizes background reading | Writes the brief, reads sources in the original |
| Drafting | Writes first drafts of descriptive passages | Rewrites, adds analysis and judgment, owns every word |
| Fact-checking | Lists every claim and flags those its source does not support | Opens every source, verifies every number, quotation and product fact |
| Editing | Suggests line edits for clarity, consistency and style | Accepts or rejects each edit, checks logic and certainty |
| Research design | Writes critiques of a study design (two models from different companies) | Answers every point in writing and decides the design |
| Analysis code | Helps write and debug code | Reviews and runs the code; reproduces every figure from raw data |
| Publishing | Prepares metadata and checks it against the text | Publishes every article |
| Social media | Prepares and schedules standard posts about published articles | Approves in advance any post about flagship research or the Index, or that names a person |
Where we never use AI
We do not use AI to:
- invent or estimate statistics, quotations, sources or interviews;
- decide the conclusions of our research or our editorial judgment;
- replace a person’s verification: a person opens the original source for every number, quotation and product fact before publication;
- appear as an author or be credited in a byline;
- create images to illustrate our articles;
- create realistic images, audio or video of real people, products, places or events;
- make charts, tables, screenshots or any image used as evidence.
The only exception is an AI system’s own image output when that output is the object of our research. We publish it as labeled evidence of what the AI produced, never as a depiction of reality (see Images).
Human responsibility
Every piece is reviewed and approved by our editor-in-chief, Nhi Phan, before publication, and the editor-in-chief publishes it. No AI system can publish to our website or send our newsletter. The person named in the byline is accountable for every word, whatever tools were used. Each article’s “How this was made” note says which steps were AI-assisted and who verified and edited the piece.
Disclosure on every article
Each article ends with a short How this was made note, for example:
How this was made: Drafted with AI assistance from a brief written by the editor. Claims checked by an AI tool, then verified against the original sources by Nhi Phan. Edited and approved by Nhi Phan.
Research reports also state which AI-assisted steps were used in data collection and analysis, and name any third-party data provider.
Confidentiality and data protection
We do not enter confidential source material, embargoed information or personal data about readers or subscribers into AI tools. We use AI services through business API settings under which, according to the providers’ terms, our inputs are not used to train their models.
Images
We do not publish AI-generated illustrations. Our images are:
- charts and diagrams we make ourselves from our data;
- screenshots we capture and document;
- photographs we own or are licensed to use, including portraits supplied by interview guests with their permission;
- abstract data art that we generate with our own code from AVJ data. No generative AI model is involved. It is credited “Data art: AVJ”, and its embedded metadata identifies it as algorithmic media (IPTC Digital Source Type “algorithmicMedia”). Data art is decorative and is never used as a chart or as evidence.
One exception. When an image produced by an AI system is itself the object of our research (for example, how an image generator depicts a brand), we may publish it as evidence. Its caption labels it as AI-generated evidence and states the model, the exact prompt and the date it was generated. We keep any provenance metadata it came with, such as C2PA Content Credentials. We never publish such an image as an illustration or present it as a real photograph.
Changes to this policy
AI tools change quickly. We review this policy at least every six months and record material changes below.
Change log
- October 22, 2026: Policy first published.
Questions: [email protected]