that 95% is doing a lot of work in there. 8,936 clicks gone and nearly all of it blog, docs and support, while the product searches held
before spending agent hours winning those back i'd want to know what they converted at last year. a docs page that answered someone's question and lost them anyway wasn't revenue then either, so recovering it might not move anything you actually care about
ran our own checkup on our own site this week. it failed in four places
no machine readable description anywhere. duplicate meta on the homepage and the agent page. lab lcp 6.4s. and an assistant, asked cold what navlio is, said we're a logistics and shipping company
which is fair. the name looks like freight and we never wrote down what we do anywhere a machine could read it
those two percentages aren't measuring the same thing though. bedrock has no reservations or oversized instances to get wrong, so most of what makes up the classic 29% can't even happen there
and you still found 7.5%, in rate errors and responses that never finished. that's a different problem from an idle instance and probably a harder one to spot
on the spend cap worry you raised with anton: for a side project, going offline is the correct failure. a cap tripping means traffic went abnormal, and abnormal on a side project is bots far more often than it's a launch
pick a number that would annoy you to pay and leave it there
@xaviervalmora the ones we've run into weren't overspending on compute at all. it was data transfer and managed services that outlived whatever they got provisioned for, and nat gateway processing charges are usually the biggest line nobody has opened
@vercel the ones that hurt are the steps bleeding in production that reproduce perfectly fine on the laptop of whoever built them. a recording of what the agent actually tried is more use there than any alert we ever wired up
loaded it, and two things come before anything about the design
over https the cert being served doesn't cover https://t.co/5BwMgEu0gA, so anyone typing the domain in or following an https link gets the full browser warning page. plain http loads fine, which is probably why nobody has mentioned it
second, the homepage ships no title tag at all. google has nothing to put in the result and an assistant asked what proofioo is has nothing to read except the domain name
the page itself is clear. those two are just sitting in front of it
@lolalallou cut myself off there, sorry. which is the fixable one. a falling click rate is the pin image and the title doing less work, not you posting less of them
1.23% overall, 4,034 outbound clicks on 327,439 impressions. worth pulling those two apart before deciding it's the algorithm. a steady rate with fewer impressions is distribution and it isn't yours to fix. a falling rate means pins are still getting shown and the image stopped earning the click, which is
@thedanielbudai discounts and paid traffic raise conversion out of the same pot the profit comes from, so the dashboard improves and the margin quietly pays for it. speed and checkout friction are the two that don't, which is the boring reason to spend the time there first
@LambdaAPI@ollieforsyth we sat on a $340 a month cluster for a side project with four users for about a year. nothing like $40k, but the same failure underneath it, nobody was checking the number monthly so it just kept renewing itself
@shipordie_@abyonm what does that hero photo weigh? just loaded the page and it's the largest element on it, which makes it the thing your lcp sits waiting on
better to know that now than when app store traffic turns up
one that's actually worth running: ask an assistant to describe the company cold, ten separate times, and count how often it names what you sell. we do this on our own site and it still misses, usually because nothing on the page states the category in plain text so it guesses off the domain
@datawithshash a real-time uptime dashboard. refreshed every few seconds, cost us money to keep fed, and nobody ever had it open
every outage we had still reached us the same way it always did, which was a customer telling us
@iamkunaldesai worth splitting that 18% by query type in search console before filing it as a win. informational queries falling while demos climb is the trade doing what you wanted
branded queries falling would be a separate problem that shows up as the same number in the chart
staging schedules save you a few hundred a month, not tens of thousands. $55k over 39 apps is about $1,400 each, so i'd go hunting for whatever is duplicated 39 times
group cost explorer by usage type rather than by service for one month. nat gateways are usually the one, about $33 a month each in us-east-1 plus $0.045 per gb through them. same for idle load balancers and gp2 volumes still attached to nothing
@JacobCounsell the bit nobody mentions is that llms.txt is a second copy of the site that nobody keeps updated. six months in it still lists last year's pricing page, and that's the copy a model quotes back at your customers
@jakezward we ran our own site through this last week and one assistant came back saying navlio is a logistics and shipping company. it's guessing off the domain, because nowhere on the page do we say the category in plain text. duller than any acronym on that whiteboard
@ihteshamali self hosting it still means paying dataforseo per pull, so you're renting the same index, just metered instead of flat
which is cheaper right up until the month you pull ranked keywords and backlinks for twenty competitors. there's no cap on that side