Strategy | Tactics

Build Your 2027 Website Plan, Live!

with the legends at Prismic: Nathalia Gaspar, Samuel Horn and Tosin Adelowo
How to build a website roadmap for 2027 Nathalia Gaspar is a web strategist at Prismic, where she runs the company’s own website and is working on turning it into an autonomous one. For the final session of the School of Webcraft, a four-week course run with Prismic, she took attendees through building a website […]

Listen on Apple Podcasts
Listen on Spotify
Listen anywhere else

How to build a website roadmap for 2027

Nathalia Gaspar is a web strategist at Prismic, where she runs the company’s own website and is working on turning it into an autonomous one. For the final session of the School of Webcraft, a four-week course run with Prismic, she took attendees through building a website roadmap for 2027, live.

In this session, Nathalia covered how to assess where your site is today, the three changes she’s planning around, a simple way to sort ideas by impact and effort, and how Prismic uses AI and MCP to update its own pages. She was joined for the Q&A by Samuel Horn af Rantzien, who looks after Prismic’s website development, and Oluwatosin (Tosin) Adelowo, who brings a growth marketing view.

Table of Contents

The short version

A good website roadmap starts with where your site is today, then sets one business goal, names the audience, and sorts every idea by impact and effort. Nathalia is planning around three shifts for 2027, which are AI search, speed, and personalisation. She also showed that AI can take over a lot of the manual website work, as long as a human approves what goes live.

How do you start a website roadmap?

Start by scoring where your website is today. Nathalia suggested rating yourself from one to five on five questions. Are the right people finding the site, or just more people? Are visitors taking the step you want, such as a demo request, a sign-up, or a form submission? Can your team launch and change pages quickly, without a developer? Who updates the content: you, AI, or a third party? And does AI search mention your website at all?

The last one is the question most people skip. We search for plenty of things, Nathalia pointed out, but rarely our own site, so try it and look at what the AI says about you. Your scores show you where to focus the plan.

In a poll of attendees, 51% named traffic that doesn’t turn into pipeline as their biggest website blocker for 2027. The other options were slow page launches, content that’s hard to keep up to date, and being hard to find in AI search.

How should you prepare your website for AI search visibility?

Check how AI assistants describe you today, because AI search is where a growing share of your visitors start. Nathalia shared that people arriving from AI search convert 4 times faster than people from regular search. They also ask assistants far more than “what’s the best CMS”, so you want to show up in Google’s AI results and in tools like ChatGPT.

Prismic has tested a few things. Adding FAQs to content improved the team’s visibility in AI tools, Nathalia said. Comparison pages are one of her quick wins for AI search. And tracking AI mentions and rankings gives you something to measure against, and Nathalia said Tosin is the person to ask about it. She called AI search the new front door of your website.

What else is changing for websites in 2027?

Speed and personalisation are the other two shifts on Nathalia’s list. Topics, content, and AI itself change week to week, so she focuses on improving website processes and automating as much as she can.

Personalisation means building pages for different audiences. That could be an account-based campaign aimed at named companies, or landing pages that change depending on where a visitor came from and how many pages they’ve viewed. All of it has to stay GDPR compliant, she added. Nathalia was clear she isn’t an oracle, so treat these as a starting list.

What goes into a website roadmap?

Nathalia’s website roadmap has five parts, and she builds them in this order.

The first is the business goal for the site. That could be more demo requests, more sign-ups, more traffic, a new market or language, faster campaigns, or showing up in AI search. Prismic’s goal is more demo requests and an autonomous website. The second is the audience. Prismic speaks to developers and marketers, and you might have one audience or five or six, each with its own strategy.

Third come the pages and structure. What does each audience need to find, how do they move through the site, do the pages deliver, and why does one page work where another doesn’t? Fourth is the content engine: who creates content, whether AI writes a first pass with human review, whether it’s all human or an agency, and how often you’ll update. Fifth is measurement, such as demo numbers and traffic, so you can show the plan is working.

The plan will change, Nathalia said, and that’s fine. It’s there to guide you.

How do you prioritise website ideas?

Sort every idea into one of four boxes by impact and effort. Nathalia does this because her favourite ideas aren’t always her best ones.

Quick wins are high impact and low effort. Her examples were refreshing the homepage hero, adding demo calls to action to the top of blog posts, fixing tracking gaps (which sometimes break), launching comparison pages, and writing FAQ content. Big bets are high impact and high effort. When you judge effort, count people as well as time: legal, developers, and AI tokens all add up. Fill-ins are low impact and low effort, so do them if you have spare time. Not now is low impact and high effort. Some of that is mandatory technical work behind the scenes, so check before you drop it.

Her list of ideas also included personalised pages for top accounts, pages tailored by industry, and testing a new pricing page, including with AI-powered A/B testing tools. On the measurement side, she suggested checking results by segment and retiring pages that don’t perform. Prismic is auditing its own site and pruning blog posts, rewriting some and removing others, and Nathalia said that’s improving its SEO.

How does Prismic use AI and MCP to run its website?

Prismic connects its website to AI tools through MCP, so Nathalia can work in ChatGPT or Claude and the changes land in Prismic. The team uses it to build and edit pages, rewrite content, analyse performance, personalise landing pages, optimise for SEO and GEO, and create images.

She timed it. A new page takes her 30 minutes by hand, and the MCP setup did it 6 times faster. Pages follow an SEO skill, a content writer skill, and Prismic’s brand guidelines, with positioning documents pulled in from Notion. The AI picks from page components called slices, and Nathalia, who has a design background, can swap any module it proposes. For images, she gives the AI a reference to copy instead of a blank brief.

Her bigger bet for 2027 is an autonomous website. She’s gathering data from Amplitude for traffic and from Google Search Console and Ahrefs for SEO and GEO, and training agents that learn from the changes she makes. The skills act as an SEO expert and a content creator to guide the machine, not to replace anyone. The missing piece, she said, was why each change was made, which usually lives only in a web manager’s head. So she and Samuel built a log where every MCP change records its intent, and the system can then flag things like a ranking drop and suggest a fix. It’s early days.

Who owns the website when five teams disagree?

Whoever has the expertise should lead, and Nathalia’s view, as a marketer, is that this is usually marketing. Her practical advice is to get one person from each team in a room, tell them you’re building the roadmap for 2027, ask for their priorities, and fit them in where they line up. It helps if sales and marketing targets are aligned first, and marketing often isn’t good at marketing itself, so being open about the plan makes it easier. Sometimes a request from high up lands on the roadmap even when it isn’t strategically right, and then you adapt.

Is it safe to let AI edit your website through MCP?

It can be, as long as the setup keeps a human between the AI and your live pages, and Samuel stressed that this depends on the MCP server you use. In Prismic’s version, the MCP puts content into a release, which is a copy of the page. The live page stays untouched until a person publishes. If you don’t like the result, you can ask again, delete the copy, or publish it. Prismic has also added undo for changes made through MCP.

There are more guardrails. The AI can only build pages from slices the team has already approved, so it can’t create a CTA in a different colour or rearrange an FAQ section, and it can archive pages but not delete them. Prismic keeps personal data out of AI tools altogether. Tosin also builds friction into his content workflow on purpose, with gates where nothing moves on until he has said he approved it. He admitted that works only if you read what you approve. Samuel’s caveat was that other MCP servers could have no gates at all, so check the implementation before you trust it.

How do you prove the ROI of AI website automation?

Don’t measure it on speed. Tosin said Prismic’s website team has no speed KPI, and nobody gets credit for publishing a page faster. Nathalia is measured on relevant web traffic and pipeline from inbound, and “fast is not always the goal.” If AI gives you worse quality than doing it yourself, he advised, work the way you did before.

Nathalia added a personal reason. She wants less time on manual jobs, like changing a logo or link on every page, and more on strategy, so she tracks how much of her time goes on small tasks. Samuel said Prismic saw the return in coding early because it’s a developer-heavy company, and the team then asked whether content could go the same way.

Where should you still rely on human judgement?

Strategy, creative work, and the final review. Samuel sees few places in website development where AI shouldn’t be used. Coding is largely handled by AI now, he said, so a developer’s work has moved to the edges: asking the AI to question you before it builds, then reviewing and running end-to-end tests. The areas he’d hold back are content and customer data, depending on the site.

Nathalia hasn’t handed conversion rate optimisation to AI yet. The test ideas she’s seen from an A/B testing tool weren’t strong, and some repeated tests she’d already run because the AI didn’t have that context. She uses AI as a starting point, not an ending point, for strategy, and she prefers human brains for creative work.

Tosin’s warning was not to let AI do your human-in-the-loop step. A workflow with an AI writer, an AI reviewer, and an AI approver can fall into the same trap. He also said not to let AI rob you of your opinions. What’s unique in content comes from your opinions, your experience, and your conversations with customers, so even adding your own view to an AI draft makes it more distinct. He doubted that any tool could research a niche, regulated industry deeply enough to find angles nobody else’s agents would.

Frequently asked questions

What should a website roadmap include?

Nathalia’s website roadmap for 2027 has five parts: a business goal for the site, the audience you’re targeting, the pages and structure each audience needs, a content engine that says who creates content and how often, and a way to measure results. Start by scoring where your site is today, and expect the plan to change.

How do you decide which website ideas to do first?

Sort every idea by impact and effort into four boxes: quick wins (high impact, low effort), big bets (high impact, high effort), fill-ins (low impact, low effort), and not now (low impact, high effort). When you judge effort, count people and approvals as well as time, such as legal, developers, and AI tokens.

Start by checking how AI assistants describe your website today. Nathalia said Prismic tested adding FAQs to its content and found it improved visibility in AI tools. She also recommends comparison pages and tracking your AI mentions and rankings. She added that visitors from AI search convert 4 times faster than those from regular search.

Is it safe to let AI edit your website?

It depends on how your MCP server is built, said Samuel. In Prismic’s version, AI changes go into a release, which is a copy of the page, and nothing goes live until a human publishes it. Changes can be undone, the AI can archive but not delete pages, and it can only use pre-approved page components.

How do you measure the ROI of using AI on your website?

Measure the goals you already had, not speed. Tosin said Prismic’s website team has no speed KPI and is judged on relevant web traffic and inbound pipeline. Quality matters most. Nathalia also tracks how much of her time goes on small manual tasks, because the aim is more time for strategy.

Where to start

Nathalia’s advice was to assess what you have and pick one goal for next year. She also suggested giving the AI your context by talking to it, which works in ChatGPT and Claude. She learned that from her mum, who is 75 and the only person Nathalia knows who uses voice search on Google.

Thank you to Nathalia, Samuel, and Tosin for being so open about how they run their own website, including the worries they had at first. Prismic is also offering a free AI search gap analysis if you’d like to see how your site shows up right here.