惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

S
SegmentFault 最新的问题
云风的 BLOG
云风的 BLOG
Threat Intelligence Blog | Flashpoint
Threat Intelligence Blog | Flashpoint
博客园_首页
CTFtime.org: upcoming CTF events
CTFtime.org: upcoming CTF events
The GitHub Blog
The GitHub Blog
Google DeepMind News
Google DeepMind News
M
MIT News - Artificial intelligence
博客园 - 叶小钗
MongoDB | Blog
MongoDB | Blog
N
News and Events Feed by Topic
Microsoft Security Blog
Microsoft Security Blog
Apple Machine Learning Research
Apple Machine Learning Research
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
T
Tailwind CSS Blog
Google DeepMind News
Google DeepMind News
IT之家
IT之家
W
WeLiveSecurity
P
Proofpoint News Feed
Exploit-DB.com RSS Feed
Exploit-DB.com RSS Feed
月光博客
月光博客
Schneier on Security
Schneier on Security
博客园 - 三生石上(FineUI控件)
Application and Cybersecurity Blog
Application and Cybersecurity Blog
腾讯CDC
H
Heimdal Security Blog
Y
Y Combinator Blog
Engineering at Meta
Engineering at Meta
量子位
宝玉的分享
宝玉的分享
博客园 - 【当耐特】
V
Visual Studio Blog
L
LangChain Blog
Last Week in AI
Last Week in AI
The Cloudflare Blog
Hacker News: Ask HN
Hacker News: Ask HN
Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
Security Archives - TechRepublic
Security Archives - TechRepublic
cs.AI updates on arXiv.org
cs.AI updates on arXiv.org
SecWiki News
SecWiki News
Simon Willison's Weblog
Simon Willison's Weblog
Security Latest
Security Latest
A
Arctic Wolf
T
Tenable Blog
I
Intezer
P
Privacy International News Feed
Attack and Defense Labs
Attack and Defense Labs
N
News | PayPal Newsroom
Martin Fowler
Martin Fowler

Amplitude

Beyond the Rate: Retail Banking's New Competitive Front How NS Prevented €1.8M in Revenue Loss Through Experimentation Go from Product Launch to Insight to Action in Minutes What Makes a Good vs Bad North Star Metric The Role of Feature Management in Successful Product Development Cohort Retention Analysis: Reduce Churn Using Customer Data 7 Steps to Measuring the Success of a Feature 14 Best Product Management Tools for 2026 (Plus Tips from Senior PMs) Putting A Number On AI Quality Meet the Winners of the 2026 Amplitude AI Impact Awards Beyond Last-Touch Attribution: Find Out Which Interactions Really Matter Agent Connectors Are Better Together Agents That Act on What Actually Happened How Square Used Amplitude to Enhance the Seller Experience and Power Growth Migrating Analytics Platforms Without The Chaos Wanted Lab Grows Sign-Ups by 150% & Builds Experimentation Culture How to Balance Inference Cost and User Experience for Agents Introducing Zoning Insights: Web Intelligence at a Glance Five best practices for getting started with AI agents 24 Quarters at #1. Here’s What’s Next. How We Built a Product That Tells Us What To Build Next: Inside Amplitude Wave Looking Beyond Campaign Metrics: 7 Marketing Success Stories AI Evals for Product Managers: A Beginner’s Guide to Getting Started The Builder Skills Library Introducing Agent Connectors in Amplitude Understand How AI Thinks, Get Better Results How We Redesigned Amplitude Docs for Agents and Made Everyone an Author AI Broke Your Experimentation Program. Here’s How to Fix It. Every Stuck User Is a Support Ticket Waiting to Happen Tracing the Sale: Connect Behavior to Conversions with Persisted Properties Building CLI Agents: It’s What You Don’t Give Them That Counts Three Tips for Better Prompts in Amplitude Global Agent How AI Took the Data Analyst’s Job, and Created a Better One Default Prompts Are Tanking Your Agent’s Retention Optimizing Core Web Vitals with Amplitude’s Global Agent Don’t Ask Global Agent Anything, Ask These Three Things How We Built a Design Agent at Amplitude with Claude Managed Agents and Cloudflare The Problem with Chasing Churn How Hostinger Achieved a 20%+ Conversion Lift Through Experimentation How STAGE Streams Smarter by Putting Data at the Center Building the Validation Stack for AI Product Development Making AI Analytics Safe for Financial Services Teams Most Teams Ship Agent Personalities by Accident. We Didn’t. What I Learned Pointing a Ralph Loop at My Product for a Week How Mercado Libre Scales Decision Making with AI Claude Cowork for PMs: 5 Playbooks to Get Started How ACKO Drove 13% More Conversions & 50% Drop in Calls with GenAI Agents Just Made Your Feature Launch Channel Smarter Homegrown FinOps Tools: How AI “Build” Beat “Buy” for Us in <1 Year Introducing The Amplitude Quickstart Series Rebuilding Session Replay’s Delivery Layer to Be Lighter on Your Page The Eval Signal That Predicts 3x Agent Retention Agents Write Code. Fixing It Is Still On You. Amplitude and Statsig Partnership 5 Agent Skills to Automate Your Weekly Product Review Amplitude Plug and Play: New AI Plugin in Claude and Cursor Marketplaces Introducing Amplitude Wizard CLI: Set Up Amplitude from Your Codebase Making AI Search Count (and Convert) How VEED Evolved Its AI Search Strategy What’s New with Amplitude Agents Effortless Support at Scale: Making Human Support More Human AI Week 2026: Upleveling All Together Amplitude AI Builders: Paul Hultgren Chats about AI Assistant Dashboard Dread to AI-Driven Decisions: How Tira Rebuilt Its Analytics Workflow Your Product Deserves a Better Support Agent How Cisco Systems Accelerated Adoption by 20% Through Data Innovation
Amplitude Heatmaps Update: More Reliable Screenshots and Accurate Placement
Chad Swenson · 2026-05-14 · via Amplitude

A heatmap is only as good as the screenshot underneath it. If the background doesn’t match your live page (elements are cut off, layout is shifted, or dynamic content is missing), the click and scroll data drawn on top of it is meaningless. You’re looking at a broken approximation of your page.

For teams running dynamic sites, that was their experience of Heatmaps. We heard it enough times that we stopped patching and started over.

When the canvas is wrong

On SPAs, Angular apps, and pages with hover states or dynamic elements, the captured screenshot often had little resemblance to the live page: elements were cut off, layouts were shifted, and content was missing entirely. The screenshot is the canvas every click dot and scroll gradient gets drawn on, so when it’s wrong, everything on top of it is wrong too. You end up staring at data you can’t trust on a page you barely recognize.

Patching individual issues wasn’t going to fix that. The problems were too interconnected—screenshot capture, element placement, device handling, URL targeting—and each fix exposed the next one. So we rebuilt the whole thing.

A screenshot that actually matches your page

There are now two ways to capture a screenshot for a heatmap: directly from your live site or from a replay. Direct capture, available now to all customers on the latest SDK, is the more significant change. It captures exact page states (hover menus open, navigation expanded, dynamic content loaded) that replay-based capture historically missed or flattened into a default view. You get the page as users actually see it, rather than a stabilized version of it at rest. We also mapped each heatmap to exactly one screenshot, removing a persistent source of confusion about which background the click data was plotted against.

Clicks that land where they should

Even when the screenshot was right, the click data placement sometimes wasn’t. Heatmaps previously positioned data using pixel coordinates, which worked well enough on static pages but broke predictably on anything that shifted between loads, breakpoints, or viewport sizes. Selector-based placement fixes this by anchoring data to the element itself, rather than its position at a specific moment. On responsive layouts and dynamic pages, the dots land where they should regardless of how the page reflows.

Less setup, more analysis

A lot of the friction in Heatmaps was everything you had to navigate before getting there. Manual device selection that produced the wrong screenshot if you guessed wrong. A URL dropdown with enough edge cases that customers had learned to route around it. A scrollmap that showed you the data but made you work to interpret it.


Device detection is now automatic, so you’re not making a choice that affects the output. The URL handling has been fixed and simplified. The scrollmap has a new design with a clearer fold line and a more readable drop-off curve, so you can see where users stop scrolling without having to decode the gradient. And the UI has been cleaned up overall, with unnecessary fields and states removed, to shorten the path between opening Heatmaps and having something worth looking at.

Heatmaps you can trust

Before this rebuild, using Heatmaps on a dynamic site meant working through a checklist of things that might be causing what you were seeing: Was the screenshot captured from the right state? Did the device type affect the layout? Are the click dots actually landing on the right elements? By the time you’d ruled those out, the analysis had become a debugging exercise.

The rebuild eliminates that checklist. Screenshot capture is more reliable, placement is anchored to elements ratheThe value of a heatmap is simple: see where users engage, spot what's working and what isn't, and make a better decision about the page. This rebuild makes sure the screenshot, the placement, and the setup are no longer what stand between you and that answer.