Framer vs custom code: which website stack AI search reads best
Benjamin Libor9 min read
A website's stack decides how much control a team has over what ChatGPT, Perplexity, Gemini and Google's AI Overviews actually read: the HTML, the structured data, the language versions and the rules for AI crawlers. Framer gets the basics right and limits the rest; a custom-code site gives full control over all of it, and with Claude doing the technical work, the marketing team runs it day to day without a developer. Neither stack guarantees rankings or citations, but custom code removes technical obstacles and makes every SEO and AEO fix cheaper to ship.
In short
- AI assistants can only cite what they can read, so control over semantic HTML, structured data, hreflang and crawler rules matters more as AI search grows.
- Framer builds in titles, sitemaps and redirects, but generates much of the HTML itself and leaves schema as manual custom code per page.
- A custom site (for example built with Astro) can add structured data automatically per page type and stay light, which in our own projects means PageSpeed scores of 95 to 100.
- For a large example site, Framer costs about $2,400 a year as of November 2026, against about $720 to $925 for custom code run with Claude.
- No stack guarantees rankings or mentions in AI answers. Content quality, authority and links still decide those.
Why does a website's tech stack matter for AI search?
When an AI assistant searches the web, it fetches pages, reads their text and decides what to quote. Pages with a clear heading structure, real lists and articles in the HTML, structured data that says what the page is, and language versions marked up correctly are easier for it to read and attribute. How assistants pick their sources is covered in how ChatGPT decides which pages to cite.
The platform decides how much of that the team controls. On a hosted builder, the platform writes the HTML and the team works inside its settings. On a custom site, every tag, schema block and crawler rule is the team's to set.
This matters most for B2B companies at Series A to B, where the site grows quickly: more languages, 15 or more CMS collections, gated content, A/B tests. That is when add-on costs climb and the ceiling on speed and SEO control starts to show. It is also why the stack should be decided early: once a site is built deep into one platform, switching means a rebuild.
How do Framer and custom code compare for SEO and AEO?
| Element | Framer | Custom code + Claude |
|---|---|---|
| Titles, descriptions, social images | Built in, per page and CMS item | Built in via templates; Claude fills them on every new page |
| Semantic HTML (headings, articles, lists) | Partly controlled, generated by Framer | Fully controlled |
| Structured data (schema.org) | Manual, custom code per page | Automatic per page type: Organization, Article, FAQ, Product |
| Sitemap and robots.txt | Automatic | Automatic, fully editable |
| Redirects | Built in on Pro | Built in, managed as a file |
| Localisation and hreflang | Automatic, not editable | Automatic, fully editable, localised URLs |
| Internal linking | Manual | Manual, or Claude suggests related links on each article |
| AI-readable content (llms.txt, AI crawler rules) | Readable, little extra control | Full control |
Three rows carry most of the weight for AI search:
- Structured data. On Framer, schema is custom code added page by page, so it tends to be missing or out of date. On a custom site, every article, product and FAQ page gets it from its template. What schema does and doesn't do is in structured data for AI search.
- Semantic HTML. A page whose headings, lists and articles are real HTML elements is easier to parse than one assembled from generic layers.
- Crawler rules. Deciding which AI crawlers may read which pages, or adding a file like
llms.txt, is a file edit on a custom site. See AI crawlers explained, and whether llms.txt matters before you spend time on it.
The bigger gain is speed of change. On a custom site, an SEO or AEO recommendation becomes a prompt, not a ticket.
Does page speed affect how AI and Google read a site?
Speed is part of how Google judges a page, through Core Web Vitals, and a light page is simply less work for any crawler to load. In our own projects, a well-built custom site (for example on Astro) realistically scores 95 to 100 on Google PageSpeed, while Framer sites are hard to push above 90 on mobile. Those are achievable results, not guarantees.
The reason is code weight. Framer ships its own JavaScript runtime on every page for animations and interactivity. A custom site ships plain HTML and adds JavaScript only where a feature needs it, which means faster loading, better Core Web Vitals and fewer layout shifts.
On either platform, the score is decided by:
- third-party scripts (analytics, cookie banners, chat widgets, tag managers), each of which can cost several points on mobile
- large hero videos and uncompressed images; custom code can resize and convert images automatically when the site is built
- fonts and heavy scroll animations
- content added later; custom code can block a publish that breaks a speed or accessibility check
Both stacks can meet WCAG 2.2 AA for accessibility. Custom code gives direct control over headings, labels, focus states, contrast and keyboard navigation, with automated checks on every change. In Framer, it depends on how each component was built.
Framer vs custom code at a glance
| Area | Framer | Custom code + Claude | What it means |
|---|---|---|---|
| Yearly running costs | From about $2,400/yr for a large site | About $720–925/yr | Save about $1,500+ per year |
| Hosting and security | Included, nothing to maintain | Cloudflare (hosting, CDN, SSL, firewall) and GitHub (history, backups). Set up once, runs itself | You own everything, no lock-in |
| Page speed and accessibility | 90+ on mobile is hard to reach and hold | 95–100 achievable | Keep third-party scripts lean |
| SEO and AEO | Basics built in, limited control of HTML and schema | Full control: schema, semantic HTML, hreflang, llms.txt | Easier for Google and AI to read. No ranking guarantees |
| Blog and content | Built-in CMS | Ask Claude, check the preview link, approve | Same effort; SEO and translation included |
| Design flexibility | Visual canvas, drag and drop | New sections and features by prompt, no platform limits | Ship new ideas faster |
| Design system | Shared components in Framer | Colours, type, spacing, components defined once; Claude only uses those | Brand stays consistent by default |
| Developer dependence | None for routine edits | None: edits, new pages and big features run through Claude | No developer needed after launch |
| Collaboration and approvals | Live co-editing, roles, staging | Every change gets a preview link and an approval; full history, one-click rollback | Name one approver in marketing |
What does each stack cost per year?
Take an example site with 3 languages, 20 CMS collections, A/B testing and 3 people editing. As of November 2026, at list prices in USD, excluding VAT and billed annually:
- Framer, about $2,400 a year: the Pro site plan ($360), 10 extra CMS collections ($480), 2 extra languages ($480), the A/B testing add-on ($600) and editor seats (about $480).
- Custom code with Claude, about $720 to $925 a year: Claude Team for 3 seats ($720), Cloudflare hosting ($0 to $60) and GitHub ($0 to $144). CMS collections, languages and A/B tests have no per-item fees.
With Framer, costs grow with every language, collection, test and seat. With custom code, they grow only with the number of people using Claude, and those seats also cover writing, research and analysis across the team. Domain costs are the same either way. Check Framer's pricing and Claude's pricing for your own numbers before deciding.
Can a marketing team run a custom-code site without a developer?
Yes. The stack is four tools: Astro for the site, GitHub for code, content and approvals, Cloudflare for hosting with a preview link for every change, and Claude Team with Claude Code, where the team asks for changes in plain language. Publishing an article looks like this:
- A marketer asks Claude to publish the article, translate it and use an attached image as the cover.
- Claude builds it from the blog template, with the SEO title, description, alt text, structured data, related links and translation.
- A private preview link shows the page within about a minute.
- The marketer asks for changes; the preview refreshes.
- The approver checks the preview and clicks Approve.
- The page goes live, and the sitemap, blog overview and language switcher update themselves.
A rules file keeps Claude to approved components and design tokens, automatic checks catch broken links, missing alt text and speed problems, and every version is kept with one-click rollback. Cloud sessions run on Anthropic's servers and connect to GitHub, so nobody installs developer tools.
When is Framer still the better choice?
Custom code with Claude fits when you want full technical SEO control, plan several languages, many pages or custom features, want to own the site with no lock-in, and already use Claude across the team.
Framer fits when a designer adjusts layouts by hand every week, you want zero technical setup and one vendor, several people need to edit the same page live at once, or the site is small enough that add-on costs stay low. The design and content work is the same either way; the difference is who controls the platform and how changes get made.
Questions people ask
Is Framer bad for SEO?
No. Framer builds in titles, descriptions, sitemaps, redirects and language versions, which covers the basics. Its limits are control: much of the HTML is generated for you, schema is manual custom code per page, and hreflang can't be edited.
Will moving to custom code get my brand cited in ChatGPT or AI Overviews?
Not by itself. No platform guarantees rankings or mentions in ChatGPT, Perplexity, Gemini or Google's AI Overviews; those depend mainly on content quality, authority and links. Custom code removes technical obstacles and makes improvements cheaper to ship.
Do I need a developer to run a custom-code website?
Not after launch. With Claude Team and Claude Code, the marketing team asks for edits, new pages and larger features in plain language, checks a preview link and approves. Name one approver in marketing so nothing goes live unchecked.
How much does Framer cost for a large B2B site?
For a site with 3 languages, 20 CMS collections, A/B testing and 3 editors, about $2,400 a year as of November 2026, at list prices billed annually. The same site on custom code with Claude costs about $720 to $925 a year.
When should we decide on our website stack?
Before you invest deeply in one platform, because switching later means a rebuild. Count your languages, CMS collections, A/B tests and editors, and run the numbers for your own site.
Written by Benjamin Libor, founder, echo.
Something wrong or out of date? Write to hello@echo-aeo.com and we'll fix it.