Website Brief Generator for Agencies: The AI Prompt That Writes Your Redesign Brief, Sitemap and 301 Map

Every web design agency has the same Tuesday. A promising lead books a call, says the site "just feels old", and asks for a price by Friday.

So somebody on the team opens a spreadsheet. They click through 140 pages, paste titles into column B, guess which ones matter, sketch a new sitemap in Figma, and hope they haven't missed the one blog post that brings in a third of the client's Google traffic.

That's 1-3 days of unbilled work before a proposal even goes out. And if they do miss that blog post, the client finds out 6 weeks after launch, when the phone stops ringing.

This post walks through a pro prompt that turns that Tuesday into a 30-minute job: a website brief generator you build once in Lovable and run before every redesign quote.

Get the full blueprint: the AI Website Redesign Brief Generator & Sitemap Planner pro prompt includes the complete master prompt (25,000+ characters), the build steps, the tool stack with real pricing, and a cold email you can send to agencies today.

Why the redesign brief is where agencies lose money

Redesigns aren't cheap. Small business redesigns commonly land somewhere between $5,000 and $25,000 with a boutique agency, and full-service shops go well past that. Paid discovery on its own runs $1,000 to $10,000 in the US, or roughly £1,500 to £5,000 for a mid-size UK project.

The trouble is most agencies don't charge for discovery on smaller jobs. They do it for free to win the work. So the brief, the sitemap and the content audit come out of margin, and they get rushed.

Rushed discovery causes 2 problems.

The first is scope. If you didn't spot the 60 thin service-area pages, the 3 duplicate "About" pages and the booking widget the client forgot to mention, your quote is wrong before you've started. You either eat the overrun or have an awkward change-request conversation in week 4.

The second one is worse: SEO. Around 60% of site migrations see measurable organic traffic loss, according to a 2025 Ahrefs analysis. A study of 892 domain migrations published on Search Engine Journal found it took an average of 523 days to get back to the old traffic level, and 17% hadn't recovered after 1,000 days. The most common cause after a redesign launch is boring and avoidable: missing or broken 301 redirects.

Agencies don't delete ranking pages on purpose. It happens because the person building the new sitemap never saw the Search Console data, or saw it in a different spreadsheet from the redirect list.

What this website brief generator does

The tool sits between "lead books a call" and "proposal goes out". A strategist feeds it 4 things:

• The client's current site URL (with a tick-box confirming the client has asked for a review and given permission to crawl).
• A goals questionnaire, filled in by the strategist or sent to the client as a link.
• Optional GA4 "Pages and screens" and Search Console "Pages" exports as CSV files.
• Up to 5 competitor URLs.

It hands back 5 things: a website redesign brief, a proposed sitemap drawn as a visual tree, a content migration map with a 301 redirect list, page-level content requirements, and a risk register. Then the client gets a private branded link where they can read it, comment on any section and click Approve.

That's the whole job. Here's how each part works under the hood.

Step 1: crawl the site properly (and politely)

The first thing the tool reads is robots.txt. Anything the site owner has blocked stays blocked, and the overview screen lists those paths as "not crawled at the site owner's request". If robots.txt blocks the lot, the tool stops and tells the strategist to ask the client for a sitemap export instead. It never tries to sneak around it.

Then it uses Firecrawl to map and crawl the site. Firecrawl renders JavaScript, returns clean main-content text, and respects robots.txt by default. Pricing is simple: about 1 credit per page, 1,000 free credits to test with, then roughly $16 a month for 5,000 credits or $83 a month for 100,000 (billed annually).

For each page the tool stores the URL, status code, title, H1, meta description, canonical, word count, page type, internal links in and out, a content fingerprint, and a last-modified date. That date comes from the XML sitemap first, then the HTTP header, then schema markup, then a visible "updated" date on the page. If none exist it says "unknown" instead of guessing, which matters more than it sounds. Older sites often have no dates at all, and a tool that invents them would flag the wrong pages as stale.

There's a crawl cap (500 pages by default, editable). Bigger sites get sampled smartly: home page, every page with analytics data, every top-level section, then an even spread. The strategist sees "sampled 500 of 3,214 URLs" in plain view and can choose to crawl the rest after seeing the credit estimate.

Step 2: flag the mess

With the inventory built, the tool flags pages:

• Thin: main content under 300 words (contact and legal pages excluded).
• Duplicate and near-duplicate: identical content fingerprints, or titles and summaries that are 80%+ similar.
• Outdated: not touched in 24 months, or still talking about "this year's prices" from 3 years ago.
• Orphaned, broken, redirect chains, missing titles, missing meta descriptions, noindex.
• Slow: mobile PageSpeed score under 50.

Every threshold is a setting. An agency that builds for law firms might want 600 words as the thin line; one that builds for restaurants might drop it to 150.

The inventory lives in a sortable table with coloured flag chips and the GA4 and Search Console numbers joined on each row. Export it to CSV and you've got the content audit appendix for your proposal, done.

Step 3: protect the pages that pay the bills

This is the bit that saves careers. When the strategist uploads the analytics exports, the tool matches them to crawled pages on URL path (GA4 gives you paths, Search Console gives you full URLs, so it normalises both).

A page gets marked protected if it's in the top 20% by sessions or by search clicks, ranks in the top 10 for any query with 50+ impressions, or has driven a conversion. Each protected page carries its reasons, like "top traffic" or "conversion page".

Protected pages can't be retired. Try it and you get a blocking warning telling you to keep, merge, rewrite or redirect it, and to give it a new URL. The share button stays greyed out until every protected row has a decision.

No analytics uploaded? Then the tool treats every indexed page as protected and shows a yellow banner saying so. Annoying, on purpose. It's the right nudge to go and ask the client for read access to their Search Console.

Step 4: performance and competitor baselines

The tool calls Google's PageSpeed Insights API for the home page and the top 5 traffic pages, on mobile and desktop. The API is free: 25,000 queries a day with a free key and no billing account. Each project uses about 12.

Results show up in plain English. "Your largest image or text block appears after 4.1 seconds; Google wants under 2.5." A practice manager can act on that sentence, and it lands far better in a proposal than "LCP 4,100ms".

Competitors get a lighter crawl (around 60 pages each) to compare structure. The output is 2 lists: "they have, you don't" (pricing page, case studies, location pages, online booking) and "you have, they don't". Those lists write half of your sales argument for you.

Step 5: the website redesign brief itself

Now the AI writes. The tool sends OpenAI's gpt-5-mini a structured summary (questionnaire answers, flag counts, protected pages with their real numbers, speed results, competitor gaps) and gets back JSON with 14 brief sections: executive summary, goals, audiences, key journeys, current site findings, what we must protect, competitor gaps, functional requirements, CMS and integrations, performance targets, content plan, timeline assumptions, out of scope, and open questions.

It never sees raw HTML, which keeps the cost low (typically under $1 per brief at about $0.25 per million input tokens). And the prompt is strict that every number must come from the data. If a figure wasn't in the inputs, it can't appear in the brief.

Every section is editable, can be regenerated on its own, and has a version history. Anything marked internal stays off the client link.

If you want to see exactly how those instructions are written, including the section list and the guardrails against invented figures, they're all inside the website redesign brief pro prompt.

Step 6: the sitemap tree and the 301 map

The AI proposes a new sitemap from the goals, audiences, journeys, current inventory and competitor gaps. The rules are tight: max 7 top-level nav items, max 3 levels deep for core pages, every key journey reachable in 2 clicks, thin pages on the same topic merged into one stronger page, and every protected page mapped to a node.

It renders as a top-down visual tree with drag-and-drop re-parenting, badges like "New" or "Merged from 3 pages", and PNG or PDF export. There's an indented list view too, because some clients hate diagrams.

Then the migration map. Every old URL gets one row with an action (keep, merge, rewrite, redirect, retire), a new URL, and a one-sentence reason. The validation catches the classic launch-day mistakes: redirect chains get flattened, loops get blocked, nothing redirects to a retired page.

Export options cover how developers actually work: Apache .htaccess, Nginx, Netlify _redirects, Vercel vercel.json, and a 2-column CSV for the WordPress Redirection plugin. Your developer pastes it in on launch day. No retyping.

Each sitemap node also gets content requirements: sections it must include, keywords (only taken from the client's real Search Console queries, never guessed), assets needed, the call to action, and who's writing it. Filter by "client" and you've got the content checklist they need to deliver, which is usually the thing that delays launch.

A worked example: Harbourside Dental

Say a 3-person agency gets an enquiry from a dental practice with 2 locations. Their WordPress site is 6 years old.

The strategist creates the project, ticks the permission box and sends the questionnaire link to the practice manager. She fills it in over lunch: primary goal is new patient bookings, audiences are nervous first-timers, families and cosmetic patients, and they need their booking software and finance calculator kept.

She also exports 12 months of Search Console "Pages" data and the GA4 pages report, because the strategist sent a 3-line how-to.

The crawl finds 212 URLs. The tool flags 47 thin pages (mostly one-paragraph treatment pages), 9 duplicates (the same "Meet the team" content on both location pages), 31 outdated blog posts from 2019 about price freezes, and a mobile performance score of 38.

Then the bit that matters. 23 pages are protected. One of them is a 2021 blog post called "How much does a dental implant cost?" that brings in 19% of all search clicks. The practice manager had no idea. In a hand-built sitemap, it's exactly the kind of page that ends up in the "old blog, retire" pile.

The proposed sitemap has 6 top-level items: Treatments, Nervous Patients, Pricing & Finance, Locations, About, Book Now. The implant post is mapped to a new "Dental implants: cost and finance" page under Pricing & Finance, with a 301 from the old URL. The 47 thin treatment pages merge into 14 stronger ones.

Competitor gaps: both nearby rivals have a dedicated nervous patients page and published price lists. This practice has neither.

The strategist spends 25 minutes editing the executive summary and fixing 4 page types, then sends the share link. The practice manager opens it that evening, leaves 2 comments on the sitemap and clicks Approve the next morning.

The agency quotes $14,500 with confidence, because the page count, content workload and migration risk are already on paper. And the approved brief becomes the scope document, so the "can we just add a few pages" conversation in week 4 has an easy answer.

AI prompts for website briefs: why one big build beats 10 small prompts

You can get a long way with single AI prompts for website briefs. Paste in a questionnaire, ask for a brief. Paste in a page list, ask for a sitemap. Lots of agencies do this already.

The gap is the data in between. A chat prompt doesn't know which URL ranks, can't crawl 300 pages, and won't refuse to delete a protected page. The value of this build is that the crawl, the analytics and the 301 map sit in one database, so the sitemap can't drift away from the redirect list.

That's why the pro prompt is a full app blueprint. The master prompt spells out the data model (16 tables with row-level security), every screen, the Edge Functions that keep API keys server-side, notifications, and a long list of edge cases: JavaScript-heavy sites, password-protected staging sites, www and non-www both live, multilingual folders, ecommerce stores with thousands of product URLs, CSVs over 50,000 rows. It even seeds a demo project so you can show the whole flow before you've crawled anything real.

The build: what you need and what it costs

The stack is small and cheap to run:

• Lovable Pro at $25 a month to build it.
• Supabase free to build and test, then $25 a month on Pro for live client use (free projects pause after a week idle).
• Firecrawl free to start, then $16 to $83 a month depending on crawl volume.
• PageSpeed Insights API free.
• OpenAI API with a spend cap, usually under $1 per brief.
• AppBuild.DIY to wrap the finished tool as a real iOS and Android app with push notifications.

The build playbook in the pro prompt is 11 steps, from confirming your Lovable plan to recording a 3-minute demo video. Expect a day, maybe 2 with testing. The longest step is running it on a couple of real sites and checking the flags against what you see with your own eyes.

That last tool is worth a word. Once the app is wrapped through AppBuild.DIY, a strategist's phone buzzes the moment a client opens the brief, comments, or clicks Approve. That's the best moment to call, and a push gets seen in minutes where an email sits until tomorrow. The Photos plugin lets them snap a whiteboard sitemap in a discovery meeting and attach it to the project too.

Who to sell it to, and the money

There are 2 ways to make money from this.

Use it in your own agency. Turn free discovery into a paid product. A "Website Redesign Brief & Migration Plan" at $1,500 to $3,500, credited against the build if the client signs, is an easy yes for a business that's nervous about losing its Google traffic. Even if you keep discovery free, you save 1-3 days per proposal and your quotes get more accurate.

Build it for other agencies. The target is a 3-20 person agency doing $5,000 to $25,000 redesigns, quoting 5+ a month. Find them on Clutch, Dribbble, the WordPress and Webflow partner directories, and in agency owner groups on LinkedIn. Ask if they've ever lost a client's rankings on a launch. Most have a story.

The suggested pricing in the blueprint:

• Setup: $2,500 to $4,500, covering the build, branding, matching their discovery questionnaire, a 3-project pilot and training.
• Monthly: $197 to $397, covering hosting, crawl and AI usage up to a cap, tuning and a monthly report.
• Optional performance bonus: $150 per signed redesign that started from a brief in the tool, capped at $750 a month.

Run the numbers on one agency. A $3,500 setup plus $297 a month is $7,064 in year one. Your running costs are roughly $25 Supabase, $16 to $83 Firecrawl, and a few dollars of AI. Sign 5 agencies and you've built a nice little recurring line from one blueprint.

And the pitch writes itself, because you can run the prospect's own website through the tool first. "I crawled your site: 184 pages, 41 thin or outdated, mobile speed score 44. Want the full brief?" is a cold email an agency owner wants to answer.

What to watch out for

It's a strong first draft. Expect strategists to adjust 10-20% of page types and merge suggestions per project, and to rewrite the executive summary in their own voice. That's fine. Editing a good draft takes 25 minutes; starting from a blank spreadsheet takes 2 days.

Without analytics exports the protection logic can't do its job, so the tool falls back to protecting everything. Make getting the GA4 and Search Console exports part of your discovery call script.

Last-modified dates are often missing on older sites, so the outdated count can come in low. The tool says how many pages had no date, so the strategist knows not to read the number as gospel.

And put crawl permission in writing. The tool records who ticked the box and when, but a line in the discovery agreement covers you properly.

Build it this week

Every redesign you quote this month will either start from a spreadsheet or from a brief that already knows which pages bring in the money. One of those takes 2 days. The other takes 30 minutes and protects your client's traffic on launch day.

Ready to build it? Get the AI Website Redesign Brief Generator & Sitemap Planner blueprint on Prompt-King. You get the full master prompt, every build step with real tool pricing, the client handover playbook, and a cold email script ready to send to your first agency.