Summarize with AI
If your conversion rate dropped after a website redesign, check three things before you blame the design: whether your tracking still counts the same event, whether the form got longer, and which sentences the new page stopped saying. Most drops come from one of those three, and each one can be fixed in a few days without rolling the site back.
This matters more for an AI startup than for most companies. A stranger decides what kind of company you are in about four seconds, and your old page may have told them in one plain sentence that nobody knew was doing the work. You know the product too well to notice when that sentence goes missing. The new page reads fine to you, because you fill the gap in your head.
The homepage that lost one sentence
Picture an AI tool that checks patients' dental insurance before their appointments, sold to dental groups with a few locations. It is sales-led: every customer starts with a demo request. The old site was two years old and looked it. The hero said AI insurance verification for dental offices. Under it sat one line: Checks every patient's plan two days before the visit and tells your front desk what is covered, what is not and what the patient will owe. Below that, a form with three fields, Work email, Office name and Number of locations, and a button that said Check tomorrow's patients.
The redesign looks like a company that has raised. Sharper type, a real product screenshot, a logo strip, dark sections that alternate with light ones. The hero now says Your front office, reimagined. The subhead says The AI operating layer for modern dental groups. The words insurance verification first appear in the fourth section, on a card titled Eligibility. The form gained Phone number, Role and a required Practice software dropdown, and the button now says Submit.
In this example, demo requests were 42 in the month before launch and 27 in the month after. The founder now opens the CRM before he opens his email. He has rewritten the new hero twice in the site editor and reverted it both times. In Monday's standup his cofounder floated going back to the old site, and nobody argued. That is the moment to slow down, because a rollback is the most expensive answer and usually the wrong one.
Check the counting before you judge the page
The first question is not why people stopped asking for demos. It is whether they stopped at all. Redesigns quietly change how a conversion gets counted. The thank-you page moves, the form tool changes, the form gets a new name, and the analytics goal still points at the old setup.
In the example, the old form sent people to a page at /thanks, and the analytics goal fired when that page loaded. The new form shows an inline message, Thanks, we will be in touch, and the URL never changes. So the analytics dashboard shows 4 conversions for the month. The real number, 27, lives only in the CRM. Half the panic in that standup came from a dashboard that stopped counting.
So count from the system that receives the lead: the CRM, the shared inbox, the booking tool. Compare the same span before and after launch, split by channel. If paid ads were paused during launch week, the mix changed and the rate will look worse for reasons that have nothing to do with the page. Then submit the form yourself, once from a phone and once from a laptop, with an email address the CRM has never seen, and confirm it lands where sales looks. If the real count did not fall, fix the event and stop here.
The demo form that grew three fields
Every new field had a good reason. Sales wanted a phone number so they could call fast. Marketing wanted the role for routing. The onboarding lead wanted to know the practice software before the call. Each request was sensible on its own. Together they turned a twenty second request into a commitment to be phoned by a stranger.
Think about who fills that form in. An office manager reading between patients who wants to see the product, not take a call during the morning rush. The phone field tells her a call is coming. The software dropdown is worse: if her system is not in the list and there is no Other option, she cannot submit at all, and no tool records that she tried. And the button changed from a promise, Check tomorrow's patients, to Submit, which tells her nothing about what happens next.
The fix keeps sales happy. Go back to the three original fields and the old button text. Ask for phone, role and practice software on the screen after the request, as optional questions, when the visitor has already said yes. Our guide to demo request page best practices covers what to put beside the form and on the screen after submit.
Read both pages side by side, word for word
Here is where most advice goes wrong. It tells you to test the new design against the old one, or to roll back and see if the number recovers. That treats the design as the thing that changed. Usually the thing that changed is a sentence, and you can find it in an hour without any test.
Pull the old homepage from your own export or an archive copy. Paste the text of both pages into two columns, top to bottom, ignoring layout. Then mark every sentence that names the category, the user, what goes in and what comes out. In the example, the old page answered all four in its first thirty words: insurance verification, the front desk, the patient's plan in, what is covered and what the patient owes out. The new page answers none of them above the fold. Your front office, reimagined could sit on twenty other sites.
Nobody cut that sentence to hurt the page. The writer wanted a cleaner hero, and the founder approved it because he reads the missing line into the new one without noticing. He knows his product, and the visitor does not. When we built the Sawdust website, a music licensing site for brands, the brief was not a look. It was to explain in one scroll why a single owner means a single license, and the page shows that structure instead of a wall of text. The look carried the sentence. It did not replace it.
So keep the new design. A site that looks like a funded company earns the visitor's first four seconds, and that is real value you paid for. Put the old sentence back under the new hero, set in the new type, and let the redesign do both jobs.
Speed and search drift more slowly
Two other causes show up later and look different. Speed first. New sites often add a video hero, a second font family and a handful of scripts. Run the old and new homepages through a page speed test on a phone setting. If the headline appears noticeably later on the new site, some mobile visitors leave before they read it, and the rate falls for everyone who arrived on a phone.
Search is the second. If total visits fell along with the rate, open Search Console. Look for old URLs returning 404, pages that ranked but now redirect to the homepage, and queries that lost clicks. Google's own guidance on site moves says to expect temporary ranking changes and that most pages of a small to medium site take a few weeks to move. It also says to keep redirects for at least a year. Notice the overlap with the lost sentence: when insurance verification left the hero, the page also got weaker for the people searching those exact words, who were the visitors most likely to ask for a demo.
If you are choosing someone to repair or rebuild the site, this ranking of website redesign agencies that protect search traffic compares studios on exactly this risk.
A one week diagnosis, in this order
- Day one: count demo requests in the CRM for the four weeks before launch and the four weeks after, split by channel.
- Day one: submit the form yourself from a phone and a laptop, with fresh email addresses, and confirm both reach sales.
- Day two: if analytics and the CRM disagree, fix the conversion event before you read any other chart.
- Day two: paste the old and new page text into two columns and mark every sentence that names the category, the user, the input and the output.
- Day three: put the strongest missing sentence back under the new hero. Do not touch the visual design.
- Day three: cut the form back to its old fields, move the new questions after submit, and restore the old button text.
- Day four: compare phone load times for the old and new homepage, if the old one is still hosted anywhere.
- Day five: check Search Console for 404s, redirects to the homepage and queries that lost clicks.
- Then leave the site alone for two full weeks of normal traffic before you judge the number again.
When rolling back is the right call
Roll back only when two things are true. You fixed the counting, restored the form and put the lost sentences back, and requests are still down two weeks later on the same traffic mix. Or the new build broke something you cannot repair quickly, such as the platform itself. Outside those cases, a rollback throws away the stature the redesign bought, returns you to the site you already decided was too small for the company, and teaches you nothing, because you changed fifty things at once, twice.
The cost of guessing is concrete. For a sales-led company the demo form is the revenue surface. Fifteen missing requests a month is fifteen fewer first calls, and the gap shows up in pipeline a quarter later. Paid traffic costs the same whether the visitor submits or not. Across all sectors the median gap from seed to Series A is around 616 days, and roughly 70 percent of seed companies never raise one. Every month spent arguing about a rollback comes out of that window.
For the next redesign, write down the sentences and form fields that do the work before anyone opens a design file. Our SaaS website redesign checklist covers the rest of what to protect before you start.
Where Studio Maydit comes in
Studio Maydit is a web and product design studio for AI founders in the US, UK and Europe, and we build in Framer, Webflow and custom code. When a redesign has cost demo requests, we start with the side by side read and the form, then make the site look like the company you have become while it still says plainly what you do in its first lines. Fixed scope work runs three to four weeks. If your numbers dipped after launch, book a 30 minute call and we will read both versions of your homepage with you.
