Muntazir

building things bots and humans can find | ex-dev, now accidentally the SEO guy | opinions are cached, not indexed

[Muntazir @cybr404]

1/ Before checking whether ChatGPT recommends your business, try one small change.

Remove your company name from the question.

It’s easy to miss why this matters. Here’s a simple example.

[Muntazir @cybr404]

@DavidGQuaid Agree, with one catch on the last line. OpenAI does say one thing: let OAI-SearchBot crawl you or you won't show up in ChatGPT search answers. Not a lot, but it's a primary source :)

[Muntazir @cybr404]

@iannuttall Ugh. If they show up in the visitor graph they're running your JS, so headless browsers, not basic bots. Search Console won't count them though. That's where you'll see if the update hit you.

[Muntazir @cybr404]

@JespernissenSEO Nice find. Noindex is a polite request, not a lock. Google and Bing honour it, other crawlers don't have to, and it looks like Grok's doesn't. I'd take Grok's own explanation of its sources with a pinch of salt though :)

[Muntazir @cybr404]

@DavidGQuaid @edwardeachday Yes, "guarantee" is the tell for me too. People who've actually run tests say "this worked for us, here's how to check it on yours". Will give the podcast a listen, thanks :)

[Muntazir @cybr404]

@JespernissenSEO Google. I can at least check a ranking. Checking citations means asking ChatGPT the same question twenty times and averaging my feelings :)

[Muntazir @cybr404]

@nathangotch Agree. My rule is do it by hand three times first. If the three runs don't match, there's nothing to automate yet.

[Muntazir @cybr404]

@joshtriedcoding Incognito won't quite settle it, same IP so you still get local results. Search Console is the cleaner check: impressions and average position turn up a day or two later, and those aren't personalised.

[Muntazir @cybr404]

@danielkbuilds Step 3 is the one I'd put a guard on. A query with impressions and no dedicated page is often already being served by a page you have, and the new one splits the signal rather than adding to it. Worth checking what ranks for it first. If it's your own page, improve that one.

[Muntazir @cybr404]

@NickLeRoy I'd change the test. Not whether the job fits in bullet points, but whether anyone would notice if the output were wrong. Which is why measurement at 5% is the line I'd argue with. Judgement is the right thing to hire for. It stays unfalsifiable until someone checks a result.

[Muntazir @cybr404]

@brodieseo Capture is the easy half. The hard one is deciding which diffs are worth showing, since most attribute changes on a live page are rotation and cache busting rather than someone making a decision. Can you mute a field in the Timeline once you know it's noisy?

[Muntazir @cybr404]

@timsoulo @ahrefs Half agree. Data is the next bottleneck, but it isn't the last one. The day you pipe in a second source you find the two of them disagree, and nothing in either dataset tells you which to show the customer. That call is the bit that still can't be generated.

[Muntazir @cybr404]

@tyd3n_ None of those, I think. You stop the day the word stops being an apology. While it's still covering for the rough edges, it's an MVP. When a bug report embarrasses you instead of flattering you, you're running a product.

[Muntazir @cybr404]

@2whazzuup Good move, though I'd drop the position filter and keep the grouping. Position is an average over impressions, so a page sitting at 9 can be top three on two queries and buried on the rest. The distinct query count is the part that's actually telling you something.

[Muntazir @cybr404]

@JespernissenSEO The tell I use is that none of them ever disagree. Agreeing is the only safe output if you didn't really read the post :)

[Muntazir @cybr404]

@tyd3n_ Nice. Worth telling your users too: App Store search reads the name, subtitle and keyword field, not the images. Screenshots are pure conversion. Which is freeing, if you think about it. The caption is the one bit of copy you get to write for a human instead of for a ranking.

[Muntazir @cybr404]

@fatjoedavies Agree, with one catch. Most help centres are a search box over a JS app, so there's no plain URL per article and nothing for a crawler to read. Check yours renders without JavaScript before you spend a month filling it. A knowledge base nobody can fetch is an internal wiki.

[Muntazir @cybr404]

@lilyraynyc Half of me wants this and half of me knows what it would show. The name often comes out of training and the link out of whatever it fetched while answering, so a lot of citations would point at a page that had nothing to do with why you got mentioned. I'd still take it.

[Muntazir @cybr404]

@Marie_Haynes Yes, Writing used to cost the sender something, so people only sent when they meant it...Now snding is free and reading still isn't. :)

[Muntazir @cybr404]

@noahkagan The bit everyone will skip is in your screenshot. Cuts capped at 5 a day, changes under 20%, cooldowns enforced. That's the actual product. An agent with no rate limit on its own decisions will walk you into a corner in a week, and the log reads fine the whole way down.