Home · Business · IT, Data & Security · Software development
As of 13 August 2026, AI can only partly improve your website's page speed.
This still needs a person who signs their name to it.
Can you do it?
15 minutesto a draft.
1 hourto something you’d act on.
Cost, all in£0
Skill neededpower-user
Who has to check ityou
What the alternative costsNo comparable human-service price is provided in the available tool data.
If this goes wrong: a rushed change can make the site slower, break a customer journey or take the site offline until you roll it back.
What to actually do
Hand it to a person
The route this page recommends
A person who owns the outcome does this end to end, worth it when the failure is dear.
Use a tool built for this
Second choiceDo it yourself
The distant thirdA chat interface, power-user skill, and roughly 1 hour until you can act on the result.
How to actually do it
- Open your website in a private browser window and record the key pages and journeys that must continue to work, such as navigation, search, forms, login and checkout.
- Run the affected pages through Google PageSpeed Insights or Lighthouse on mobile and desktop, then save the reports and note the page URL, test conditions and reported opportunities.
- Open your code repository, CMS, hosting panel or deployment configuration and gather the relevant HTML, CSS, JavaScript, image-loading code and build settings, removing passwords, tokens and personal data.
- Paste the URL, platform details, reports and relevant code into the prompt, then ask the model for a prioritised plan with exact changes, risks, tests and rollback instructions.
- Apply one low-risk change at a time in a staging environment or a reversible branch, and record the previous version of every edited file or setting.
- Run the same PageSpeed or Lighthouse tests again and compare the results with the original reports, then test the key journeys on mobile and desktop before sending the change to production.
- Deploy only changes that improve or preserve performance without breaking layout, forms, tracking, security or accessibility, and monitor the live pages before applying the next change.
Prompt
Act as a web performance engineer. Help me improve this website without inventing facts or recommending changes that cannot be tested. Website URL: [URL] Platform or framework: [for example WordPress, Shopify, React or unknown] Hosting and CDN: [details or unknown] Main goal: [for example better mobile performance, faster checkout or improved Core Web Vitals] Paste the latest results from Lighthouse or PageSpeed Insights for mobile and desktop below: [REPORTS] Paste any relevant HTML, CSS, JavaScript, server configuration or build configuration below. Remove passwords, API keys, tokens and personal data: [CODE_OR_CONFIGURATION] Produce: 1. A table of the issues, their likely causes, the evidence in the reports, and the expected risk of each proposed change. 2. A prioritised plan that starts with changes that are low risk and reversible. 3. Exact code or configuration changes only where the supplied material supports them. Mark anything that needs confirmation instead of guessing. 4. A separate list of changes that require access to the CMS, hosting, CDN, database or third-party services. 5. A test plan covering page speed, visual layout, forms, checkout or other important user journeys, mobile and desktop browsers, caching, analytics and accessibility. 6. A rollback plan for every change. Do not claim that a change will improve a score unless a before-and-after test can establish it. Do not remove tracking, security controls, accessibility features or essential functionality merely to improve performance. End by listing the exact information or access still needed before implementation.
Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.
What it gets wrong
- AI cannot see private hosting, CDN, CMS or deployment settings unless you supply the relevant information or access through an appropriate tool.
- AI cannot know which performance trade-offs are acceptable for your business, such as reducing image quality, delaying analytics or changing third-party scripts.
- AI cannot establish that a suggested fix works on your real traffic, device mix and browser mix without you running controlled tests.
- AI cannot take responsibility for an outage, broken checkout, lost tracking data or a security weakness caused by its recommendation.
- AI can produce a plausible configuration that conflicts with your existing caching, build or server setup, so every change needs a reversible implementation and test.
What caps this at PARTLY: verification cost, judgement under ambiguity and stakes of error.
How we scored this
Five axes, each scored nought to two by hand: ten means AI carries the task cleanly, and the thresholds that turn a total into YES, PARTLY or NO are published in the methodology. Each axis name links to its definition.
| Axis | Score (0–2) |
|---|---|
| Output | 2 |
| Inputs | 1 |
| Verification | 1 |
| Liability | 1 |
| Effort delta | 2 |
| Total | 7 / 10 |
The methodology and its thresholds are published in full.
FAQ
- Can ChatGPT make my website faster?
- Partly. It can interpret PageSpeed or Lighthouse reports, identify likely causes and draft code or configuration changes, but you must provide the relevant site information and test each change on your own website.
- What should I give AI to improve my website speed?
- Give it the page URL, the platform and hosting details, mobile and desktop performance reports, and the relevant HTML, CSS, JavaScript or server configuration. Remove passwords, tokens, API keys and personal data before pasting anything.
- Can AI fix Core Web Vitals?
- It can suggest fixes for issues such as large images, render-blocking resources, excessive JavaScript and slow server responses when the reports provide supporting evidence. It cannot confirm the fix without repeat tests and checks that the site still works across important devices and user journeys.
- Is it safe to let AI change my website code?
- Use AI for proposed changes rather than unrestricted production access. Apply one reversible change in staging, compare the new performance results with the original reports, and test forms, checkout, tracking, security and accessibility before deployment.
Nearby answers
Assessed by gpt-5.6-luna (gpt-5.6-luna) on 2026-08-13, second-checked by an independent model. Wrong somewhere? Email [email protected] and it gets re-checked.
The newsletter
AI news, new answers and product picks, straight to your inbox.