Feature Upvote fixed a wrong Google AI Overview in under an hour

Semrush ran a case study on its blog about how Feature Upvote, a feedback-management product, convinced Google's AI Overview to stop telling potential buyers that the company had no public REST API. The founder, Steve McLeod, is quoted on the record about the sequence that worked.

The wrong answer

A user asked Google whether Feature Upvote had an API. The AI Overview answered: "Feature Upvote doesn't have a REST API." The company had launched a public REST API the week before, but Google's answer rested on older blog posts that still described the product without one. McLeod is quoted in the piece: "Google's AI Overviews cited this much older information as opposed to our new help documentation that had Public REST API as its H1 heading."

The business cost of that answer sat on every search a prospect ran: the first thing a buyer saw about the product's integration surface was that it had none.

The three-step fix

Feature Upvote did three things, in order.

First, the team updated the older blog posts that contradicted the new API. These were the pages Google was already quoting and had the strongest ranking signals, so changing their text changed what the model would find on a fresh read.

Second, the team wrote dedicated help documentation and gave the page Public REST API as its H1. The intent was to make the strongest piece of evidence about the API the one Google saw first, with the question's own words in the biggest heading on the page.

Third, the team submitted the updated and new pages to Google Search Console for indexing by hand, through the URL Inspection tool. One hour after that manual submission on 28 July 2026, the AI Overview was returning the correct answer.

What a working team should take from this

The pattern is reusable. Three conditions have to hold at once. The old content that was wrong has to be rewritten, because an answer engine will keep quoting it. A new, highly structured page has to exist with the right question as its H1. And the pages have to be submitted for a fresh crawl rather than left to Google's own schedule.

Semrush's piece does not share impression or click numbers for Feature Upvote before or after the fix. McLeod names the time to correct the answer, not the business impact. So the method is well-sourced. The size of the gain for your own product has to be measured on your own site.

Two things that would strengthen a repeat of this. First, pick the exact query a buyer is likely to run and build the help page around that question in the voice of its words. Second, check what the AI answer says one week later, because an index refresh can revert the state if the old pages are not down for good.

For any team watching an AI Overview misstate what its product does, this is the first-hand account of the shortest path to a correction on record.

Source

Semrush: How Feature Upvote fixed Google's wrong AI Overview answer.