Effective: October 22, 2026 · Last updated: October 22, 2026
These standards govern everything AI Visibility Journal (AVJ) publishes. They exist so that readers, and the people and systems that cite us, can judge how much weight to give our work.
Scope. These standards apply to every format we publish: research, the AVJ Citation Index, guides, analysis, news, interviews, the newsletter and our official social media posts. Where a format has extra rules, they are set out below. Our AI Use Policy, Methodology and Corrections policy form part of these standards.
Accountability. The editor-in-chief, Nhi Phan, is responsible for these standards and approves everything we publish.
1. Accuracy comes first
We would rather publish later than publish something wrong. Every factual claim must be supported by a source we have read, data we have collected, or reporting we can stand behind. If a claim cannot be verified, it does not run.
2. Sources and how we grade them
| Grade | What it is | How we use it |
|---|---|---|
| Primary data | AVJ’s own measurements, with published methodology | May support headline conclusions |
| Official documentation | Platform documentation (for example from Google, Microsoft, OpenAI, Anthropic, Perplexity), laws and regulations, peer-reviewed research | Used directly; quoted where wording matters |
| Published studies | Research by organizations that disclose their methods, including vendors | Used with the method and sample described; vendor research is labeled as such |
| Expert and trade commentary | Named practitioners, trade publications | Context and perspective, never sole evidence |
| Unverified material | Social posts, forums, anonymous content, AI-generated text | Only as an example of what someone said, never as evidence |
An answer produced by an AI chatbot is never treated as a source of fact. When we report what an AI system said, we present it as an observation of that system’s output, with the date, the system and, where relevant, the prompt.
Important claims need at least one primary or official source, or two independent published studies. News stories are published only when we have opened at least one primary or official source ourselves.
3. Fact-checking
AVJ is a small publication. Every article goes through the same checks before publication:
- Claim log. Every factual claim is listed: numbers, quotations, dates, definitions, product facts and attributions.
- AI-assisted check. An AI model from a different company than the one used in drafting compares each claim with its cited source and flags anything unsupported, misquoted or out of date, and any broken link.
- Human verification. The editor-in-chief opens the original source for every number, quotation and product fact, marks each claim as verified or not, and archives the source. A single unresolved claim stops publication; it is fixed, rewritten or removed.
- Editing. The editor-in-chief checks that the conclusions follow from the evidence and that our language matches our level of certainty. AI-suggested line edits are accepted or rejected by the editor.
- Pre-publication check. Title, summary, structured data and the “How this was made” note are checked against the final text.
Because the same person often writes and verifies, the “How this was made” note on each article states who verified it. If we add an independent fact-checker or second editor, we will say so here and on the articles they check.
Original research. Before a study is published, its analysis code is re-run from the raw data and must reproduce every published figure. The design is reviewed internally before data collection, using written reviews from AI models by two different companies, which the editor-in-chief answers point by point. Where a study is pre-registered, we say “Pre-registered on OSF; design checked internally” and link to the registration. We do not describe our work as independently or peer reviewed unless it has been, and we will name the reviewers when it is.
4. Language that reflects certainty
- “We found”: a result from our own data.
- “The data suggests”: a pattern or correlation, not proof of cause.
- “We believe”: editorial judgment, only in pieces labeled Analysis.
- “It is unclear”: the evidence is insufficient.
We do not describe correlations as causes. We do not use “guaranteed”, “secret” or similar language about outcomes no one can promise.
5. Labels
Every piece carries one label, shown above the headline:
| Label | What it means |
|---|---|
| Research | Original data collected or analyzed by AVJ, with a methodology section, limitations and, where possible, open data |
| Guide | An evergreen explanation built on documented evidence, reviewed on a schedule and dated |
| Analysis | Interpretation and editorial judgment, clearly separated from what the evidence shows |
| News | A report of a recent, verifiable change, based on at least one primary or official source |
| Interview | A written Q&A with a named guest, edited for length and clarity |
| Sponsored | Paid content. We do not publish any at present; if we ever do, it will be labeled at the top, never presented as research and never part of the Index |
6. Headlines and summaries
Headlines and summaries must be supported by the body of the piece. We do not use a number in a headline that is not in the text with its source, and we do not use question headlines to imply claims we cannot support.
7. Numbers and charts
- Every statistic includes its source, sample size (n) and time frame.
- Chart titles state the finding; captions state the source, n and dates.
- We publish the data behind our charts wherever licensing allows.
- Each study names its data sources and any third-party data provider in its methodology.
- We link to every source we rely on and keep an archived copy of it, so a claim can still be checked if the original page changes.
8. Dates and updates
Every article shows when it was first published and when it was last substantively updated. Guides are reviewed for accuracy on a rolling schedule set at our monthly content review, and each guide shows the date it was last reviewed. Material changes are recorded in a changelog at the end of the article. Errors are handled under our Corrections policy.
9. Bylines and accountability
Every article carries the name of the person responsible for it. AI tools are never credited as authors. Our AI Use Policy explains how we use them.
10. Interviews and quotations
Quotations are verbatim, edited only for length or clarity without changing meaning. Our interviews are conducted in writing, by email: we send questions, the guest answers in writing, and we edit for length and clarity. Guests may check their edited answers for accuracy before publication, but they do not approve the article. We grant anonymity only when a source faces genuine risk, and we explain why in the article.
11. Screenshots and AI outputs
When we show output from an AI system, we record the system, the date (UTC), the location and settings where known, and the prompt. Because AI outputs vary between runs, we treat a single screenshot as an illustration, not evidence, unless it is part of a documented sample.
12. Independence and fair treatment
We give people and companies we criticize a fair opportunity to respond before publication, and we report their response accurately. Our commercial relationships never influence our coverage. See Ethics & Independence and Disclosures.
13. Feedback
Tell us when you think we are wrong: [email protected]. We read everything, and we respond to substantive concerns.