Catch AI models spreading false claims about your brand

Every weekday at 8am, watch for AI models stating things about my brand that are simply not true, and get my team on it before the claim spreads.

Start by pulling the most frequent false claims from the last 7 days with Profound's Accuracy Top Inaccurate Claims report, and pull the ranked themes and their response share with Accuracy Inaccurate Themes. For each claim cluster worth reporting, use Accuracy Claim Citations to fetch the specific URLs backing that cluster. If the claim data does not already tell me which AI models are repeating a claim, use Accuracy Inaccurate Clusters to get the per model breakdown.

Surface only claims that are new since the previous run or clearly rising in frequency. This filter is the difference between a useful alert and a daily wall of the same five claims, so apply it strictly. Establish it two ways. First, run the same accuracy reports over the previous 7 day window and compare frequencies, so a claim whose count is climbing counts as rising. Second, query my Notion tracker with Query a Data Source to see which claims have already been logged, so anything already recorded and flat is skipped. If nothing is new or rising, post a short note saying the check ran and found nothing new rather than staying silent.

For each surfaced claim, judge where the fix belongs and say so explicitly. Pick one of three routes: a page update on our own site when the claim traces back to our marketing pages or to content we never wrote, a correction in our documentation when the model is reading our docs and drawing the wrong conclusion, or outreach to the third party source when an external site is feeding the model bad information. Base this on the citing URLs rather than guesswork, and name the specific page or domain that needs to change.

Post an alert to my Slack channel with Send a Message. For each claim, state it in plain language the way a person would say it, then give which models repeat it, how often it came up in the window, whether it is new or rising, the citing URLs, and the recommended fix route. Lead with the highest frequency claims and keep the message scannable.

Log every surfaced claim as a row in my Notion tracker using Create a Page as a child of the tracker database. Capture the claim, its theme, the affected models, the source URLs, the recommended correction, the date it was first seen, and a status starting at open, so we keep a running record of what we fixed and what is still wrong.

Two rules to hold to. If Profound returns no accuracy data for the window, say so plainly instead of inventing claims. And never restate a false claim as though it were true in the Slack message or the Notion row, always frame it as a claim a model is making about us.