Know which changes to your site actually worked.

No. A rank tracker tells you a number moved. Growth Radar records the change you deliberately shipped, the metric it was supposed to move, and the value at ship time, so it can tell you whether the work paid off, per kind of change, across every site you own.

Sign in to Google from the connect screen and grant read-only Search Console access. Growth Radar lists the properties that Google account can see, you pick the ones to track, and the last 28 days import straight away. GA4 is a separate, optional consent from the same Google account. Disconnecting revokes the grant at Google. If Google access is not switched on yet, the connect screen says so and Bing still works. Search Console data trails real time by two to three days.

Paste a Bing Webmaster API key (in Bing Webmaster Tools: Settings, then API access). Bing has no API that issues the key, so it has to be pasted. Growth Radar checks the key against Bing before storing it, then lists your Bing sites to pick from. A site not on Bing yet can be imported there from Search Console. Several AI assistants retrieve from Bing's index, so it matters beyond Bing itself.

Yes. Every Monday morning, in your time zone, a digest of what changed on each site, ranked by the clicks at stake, with the fixes to make next. A sharp drop in clicks sends its own alert. You choose the recipients and can unsubscribe from either in one click.

It reads your homepage, robots.txt, llms.txt and sitemap the way an AI crawler does, as served and without running JavaScript: whether GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended may read the site, which JSON-LD types are present, and whether the title, description, canonical and Open Graph tags are set. It does not see your rankings, your clicks or whether AI answers already cite you. That needs Search Console and Bing, which is what the product connects to.

When a fix is marked shipped, or its pull request merges, it records the metric the fix was meant to move (clicks, CTR, position, impressions, broken links or AI citations) for the pages it touched, over the 28 days before. Until 28 days have passed the verdict is too early. Then it measures the same metric over the same pages again, and over the most similar untouched pages of the same site. The change is a win when it beats at least 95% of them and moves at least 10% more than they typically do, a loss in the mirror case, and flat otherwise. When the test was too weak to tell a real change from noise, which on low-traffic pages is the usual result, or there was nothing comparable to check against, the verdict is inconclusive and counts nowhere. With nothing to measure it is no data, never a loss. For position, load time, errors and broken links, lower counts as better. A fix whose starting number was typed in by hand rather than measured is kept out of the headline counts.

In four ways, all narrower than a dedicated AI-visibility tracker. Bing's index is what several AI assistants read, so it reads Bing's own report of citations in AI answers. Once a week it asks a web-grounded model (Perplexity Sonar by default, through OpenRouter) the questions your site should answer and records which sites it cites. It reports GA4's AI-assistant traffic, which undercounts because most assistant visits arrive without a referrer. And it checks that each site serves llms.txt, an IndexNow key, and a robots.txt that lets AI crawlers in. A fix aimed at AI citations is graded the same way as one aimed at clicks.

The aggregate reports each source returns: clicks, impressions, click-through rate and position per query and page, Core Web Vitals, and visit counts. No other account can see a site's data; every page and API call is scoped to the signed-in account, and the operator can access the database to run and repair the service. Google tokens and pasted API keys are encrypted with AES-256-GCM before they are stored, and are never shown again. To write findings, a site's top queries and aggregate numbers are sent to a model provider.

Only for the shipping half; measuring works without it. Shipping uses a fine-grained GitHub token with Contents and Pull requests write access, scoped to chosen repositories. Every change passes an allowlist of paths for its kind of fix, and anything touching authentication, payments, CI configuration, dependencies or secrets is refused outright. Growth Radar opens pull requests. It cannot merge them.

Invites go out in line order. Each person who joins with your link and confirms their email moves you up 5 places (aliases of one mailbox count once; at most 2 per company domain, 5 a day and 20 in all). An unconfirmed signup does not count, and neither does your own address.

Yes, during early access. Join the waitlist with an email address; no credit card is needed, and accounts open by invite. A paid plan will come later, and its price will be published on the pricing page before anyone is charged.

Ohad Fisher, building it because he needed it to see which of his own site changes actually paid off. The example sites and numbers on this page are illustrative.