AI providers release new versions of their models regularly, and Gumshoe moves to the new version so your Visibility Audit (formerly known as a report) keeps measuring what people are actually using. When that happens, your numbers can move even though nothing changed on your site and nothing changed in your market. Here is why, and how to tell the difference.
What actually changes when a model is updated?
More than you might expect. A new version can weight sources differently, rank brands differently, and, importantly, name a different number of brands in a single answer. That last one is the least obvious and often the biggest driver of a score change.
How can a model update move my score if nothing changed on my site?
Your visibility is the share of responses that mention your brand. So if a model starts giving longer answers that name more brands, more brands get mentioned, and visibility rises across the board without anyone publishing anything new.
The effect is uneven, and the pattern is worth knowing:
- Brands just outside the old shortlist gain the most. They were close to being named, and a longer list reaches them.
- Those new mentions usually land low in the answer, not at the top.
- Brands already named in nearly every answer change very little, because they had nowhere to gain.
The reverse happens too. A model version that gives shorter, tighter answers names fewer brands, and visibility can fall for everyone in the category at once.
How do I tell a model update from a real market change?
Model updates are marked as notes in your Visibility Audit graph.
Compare the models inside your own audit. That is the most reliable check you can run, and you already have everything you need for it.
- If one model moved sharply and the others stayed flat, the cause is that model, not your market.
- If every model moved in the same direction, something real changed: your content, a competitor's content, or how the web is covering your category.
- Check citations. A model answering from training data alone returns no source URLs, so it cannot have picked up anything newly published. A change there is model behavior by definition.
Did my competitors actually gain ground, or is it just longer answers?
It can be both, and they are worth separating. Look at rank, not just whether a brand appeared.
- A competitor appearing near the bottom of a longer answer has gained a mention, not a recommendation.
- A competitor moving into the top few positions much more often is a real change in how the model regards them, and that is the one worth responding to.
Your Competitive Leaderboard and Trends show movement run over run, so you can see which of the two you are looking at.
How do I find out why the change happened?
Your Visibility Audit tells you that something moved. Your other audits tell you why, and they are most useful worked in order rather than dipped into at random.
- Sentiment Audit: start here. It tells you whether the models' actual opinion of you changed, or only how many brands they listed. If your visibility rose but sentiment held steady, you are being named more often rather than regarded more highly, and those are two very different situations.
- Technical Audit: check whether the pages you want cited can actually be read once a model fetches them. A model naming you at the bottom of a longer answer is an opening, and readable pages are what let you move up from there.
- Content Audit: find what to publish to hold a gain or recover a loss, based on the topics and questions where you are thin.
- Citation Audit: see which pages the models are drawing from. If the sources behind your category shifted, something real changed. If the sources look the same and only the answers moved, it is the model.
Sentiment Audit, Content Audit, and Citation Audit are included with the Pro subscription. On Starter you can still compare models inside your Visibility Audit and review your Citations to see which URLs the models cited, which will tell you whether new web content was involved.
Which models does my audit use, and do they search the web?
Open your project settings and find Model families under Audit settings. Each model has a help icon, and the tooltip tells you what that model is, who typically uses it, and whether Live Web Search is on or off for it.
Check there rather than relying on a list in an article. Providers release new versions often, and the model list changes with them, so the tooltip is always the current answer.
Should I be worried when this happens?
No. Watching how model behavior shifts is a large part of what Gumshoe is for. A model update that changes how brands get named is exactly the kind of change that would otherwise be invisible to you until it showed up in your pipeline.
What you should not do is rewrite your strategy off a single run, or act on a number without knowing what drove it. Look at the trend across runs, look at rank rather than presence alone, check whether the movement is isolated to one model, and then work through your other audits to find the cause before you decide what to change.
Pro tip: If you gained visibility but the new mentions are landing near the bottom of answers, treat that as an opening rather than a win. Being named at all means the model now considers you part of the category, and moving up the list is a far shorter distance than getting on it in the first place. Your Sentiment Audit and Content Audit are where you find out what would move you up.