PageSpeed Insights Review (2026): It Found a Real Layout Bug
We picked Google’s PageSpeed Insights specifically because of a pattern across four other tools we’d tested: Surfer SEO’s Auto-Optimize button, SE Ranking’s Site Audit detail, Semrush’s Core Web Vitals widget, and Headline Studio’s SEO Score all paywall real audit data behind a paid plan. Semrush’s is the one that’s specifically about Core Web Vitals - the other three gate different kinds of detail. PageSpeed Insights is free, official, and needs no signup at all - no Phase 1 money check required for once, and it gives you real Core Web Vitals numbers with nothing locked.
(Correction, Aug 30: an earlier version of this review and its IndieHackers writeup described all four tools as paywalling “Core Web Vitals” specifically. Only Semrush’s paywall is actually about Core Web Vitals - the other three gate unrelated audit features. Caught after a reader asked a precise question about it; the broader “paywalled audit detail” pattern still holds across all four, the CWV-specific framing didn’t.)
A new category we hadn’t seen before: Agentic Browsing
Alongside the familiar Performance/Accessibility/Best Practices/SEO scores, our run surfaced a fifth category
we hadn’t encountered in any tool tested so far: Agentic Browsing, described as checks that “ensure
high-quality, browsable websites for AI agents and validate the correctness of WebMCP integrations” - explicitly
labeled “still under development and subject to change.” Our homepage scored 3/3, with llms.txt follows recommendations as one of the three passed checks - the file we added back on Aug 20 to close a gap an older
Frase audit had flagged, now getting picked up by a completely different, newer Google feature.
It found a real layout-shift bug on our own review page
Running the same check against our Semrush review page told a different story: Agentic Browsing dropped to 2/3, with Cumulative Layout Shift flagged at 0.154 (orange - “needs improvement”).
The cause: every review page embeds screenshots as plain markdown images, which render as <img> tags with no
width/height attributes - so the browser can’t reserve space for them before they load, and the page jumps
as each screenshot pops in. We checked, and this wasn’t a one-page problem: 29 images across all 10 published
reviews had the same gap.
We wrote a one-off script to read each screenshot’s real pixel dimensions and rewrite every markdown image into an explicit HTML image tag with its real dimensions attached, fixing the root cause site-wide instead of patching the one page the tool happened to flag. Re-running PageSpeed Insights against the same Semrush review page afterward confirmed the fix: CLS dropped from 0.154 to 0, and Agentic Browsing went back to 3/3.
The rest of the scores
Outside of that one bug, the site scored close to a perfect line across the board: Performance 99-100, Accessibility 100, Best Practices 100, SEO 100, on both mobile and desktop, with Largest Contentful Paint under 2 seconds even on throttled mobile emulation. Nothing else needed fixing.
One real friction point: the tool itself can be slow or stall
The web UI at pagespeed.web.dev occasionally sat on “Analyzing” far longer than expected during testing, and
the unauthenticated public API (pagespeedonline.googleapis.com) returned a flat 429 Quota exceeded with a
documented daily limit of zero queries per project without an API key - worth knowing if you’re planning to
script this rather than use the web UI, since “no signup required” only holds for the manual web tool, not
programmatic access.
Bottom line
For a site that already avoids screenshot-heavy paywalled dashboards, PageSpeed Insights is worth running after every batch of new content - it’s free, it needs nothing from you, and in our case it caught a real, fixable bug that had been sitting on 10 published pages unnoticed. The new Agentic Browsing category is still labeled experimental, but it’s a genuinely new lens (llms.txt validation, WebMCP checks) none of the paid tools we’ve tested offer yet. No affiliate program exists for it, same as Screaming Frog and Headline Studio - this is a pure usefulness recommendation.