Can You Trust AI Chatbots When Seeking Help With Databases?

In early 2023, the rapid rise of generative AI models like ChatGPT sparked widespread experimentation, prompting the author to test how well AI chatbots handle database‑related questions, particularly downgrade scenarios for MongoDB and MySQL.

The author first queried Google Bard about why MongoDB 7.0 supposedly makes downgrades easier. Bard produced a lengthy, confidently wrong answer that listed simplified downgrade workflows, enhanced compatibility, and new commands, none of which reflect the reality that binary downgrades are no longer supported in MongoDB 7.0.

When the author pressed Bard on the correct procedure—downgrading a standalone MongoDB server from 7.0 to 6.0—Bard suggested downloading the 6.0 binary, stopping the service with mongod -shutdown , replacing binaries, setting the featureCompatibilityVersion to "5.0" via mongosh -u root <database_name>\n db.adminCommand({ setFeatureCompatibilityVersion: "5.0" }) , and restarting with mongod -f <path_to_mongodb_6.0_config_file> . The response ignored the official statement that binary downgrades are unsupported and recommended contacting MongoDB support.

Next, the author asked Bard how to downgrade MySQL 8.0.35 to 5.7.44. Bard produced a step‑by‑step guide that mixed Windows and Linux commands, omitted essential backups, and suggested unsafe actions such as deleting data directories without prior backup, illustrating the danger of blindly following AI‑generated instructions.

The author then turned to Bing Chat, which correctly cited MySQL documentation that downgrading from 8.0 to 5.7 is not supported and that the only safe path is to restore from a pre‑upgrade backup. Bing also clarified that while data can be exported from 8.0 and imported into 5.7, the process is non‑trivial and requires manual intervention.

Finally, the author queried ChatGPT (3.5) about Percona XtraBackup 8.0.33, receiving a factual limitation notice about the model’s knowledge cutoff, reinforcing that AI answers can vary widely depending on the underlying model and its training data.

The article concludes that while generative AI can provide useful information, its answers to technical database questions are often inaccurate, incomplete, or even hazardous, and therefore users must verify AI‑generated guidance against official documentation and best practices.