The rename shipped 6 weeks ago. The homepage says Meridian, the launch post says Meridian, and every week another customer screenshots the old name anyway: a help article that still walks through Atlas settings, a deal in the CRM called "Atlas upsell", a meta description nobody has opened since the page was built. You fix each one as it surfaces, and each one costs the same small flinch. If that was still live, what else is?
The reason it drags is structural. The old name lives in 5 systems, and each system's search sees only itself. The site search does not read the help center, the help center search does not read the CRM, and nothing reads the email templates. The fix is to make the stack one searchable thing: pull every platform into folders on your laptop, search the old name once across all of them, let your AI agent repair every hit with the grammar intact, and review the whole rename as one queue of diffs before any of it ships.
Your options
Per-platform find-and-replace tools
Help centers, stores, and CRMs have each grown an add-on for this: find-and-replace apps with whole-word matching, draft modes, sometimes a dry run. Inside their one platform they are fast, and the better ones show you changes before committing, which is the right instinct. The limits are three. They match strings, not meaning, so "Atlas Pro" and "Atlas-powered" need separate passes and the grammar around each swap is your problem. Each covers exactly 1 platform, so the rename becomes 5 tools with 5 matching behaviors and 5 review screens. And some platforms never built one: Webflow's find-and-replace request has sat on its public wishlist since 2019, answered so far only by marketplace add-ons.
A CSV export per platform
Export, find-and-replace in the spreadsheet, re-import. It is free, it scales, and the spreadsheet is a crude review: you can read the column before you upload it. The edges are per-platform and sharp. Rich text travels badly through cells, every importer has its own rules about IDs and blank values, and the import writes straight into the platform with no per-row undo. Not everything round-trips at all: Intercom's own answer to bulk article edits is the API, one article per call, not a spreadsheet. Five platforms means 5 exports, 5 sets of quirks, and 5 chances to learn one of them the hard way.
An agency rebrand checklist
The checklist is the grown-up version of this job, and the good ones are genuinely good: a couple dozen items covering domains, legal, social handles, collateral, and redirects, with an owner and a due date per line. Treating one-to-one redirect mapping as launch readiness is advice worth taking. The limit is altitude. A checklist finds categories, not occurrences. "Update the help center" is one line; the 60 articles behind it are still one screen at a time. And the checklist ends at launch, while the old name does not.
An AI agent on per-platform MCP servers
Wire your agent to each platform's MCP server and it can genuinely search, judge, and edit. The constraint is that each MCP sees one platform, through that platform's API. Webflow's data API has no full-text search across collection items, so "find the old name" means listing everything, 100 items per request under a cap of 60 requests a minute on most plans, and filtering on the agent's side. A 3,000-item site is 30 calls before the agent has read anything once, and every fix after that is its own write call. Multiply by 5 platforms, 5 servers, 5 auth setups. And every fix writes straight into the platform the moment it runs, with no diff and no queue.
Scratch, one project, every folder
Scratch pulls each platform into its own folder inside one project: Shopify products and articles, Webflow CMS items and page metadata, Intercom help articles, HubSpot deals, companies, and notes, Brevo email templates. The stack becomes a folder of folders, and one grep, the one-line search that reads every file at once, returns every occurrence of the old name with a count per folder, in seconds. Your agent fixes each hit with judgment, and everything lands in one review queue: word-level diffs, approve per item, revert per item after publish. The cost is stated plainly: the review is real human reading, on purpose. A rename that touches 400 items is an afternoon with the diffs.
| Option | Sees every platform | Fixes with judgment | One review queue | Undo after publish |
|---|---|---|---|---|
| Per-platform find-and-replace | No, 1 each | No, literal strings | No, 1 per tool | Varies |
| CSV round trips | One export at a time | Bring your own | The spreadsheet, per platform | Your backup, if you kept one |
| Agency checklist | Categories, not occurrences | Human, line by line | The checklist itself | No |
| Agent on per-platform MCPs | One MCP at a time | Yes | Only if you build it | No |
| Scratch | Yes, one grep | Yes, your agent | Yes, every diff | Per item, even after publish |
How the loop works on a rename
- Scratch pulls your whole stack into one project. Each source becomes its own folder of local files: one per Shopify product or article, one per Webflow CMS item, one per Intercom article, one per HubSpot deal, company, or note, one per Brevo template. Then run the audit every checklist asks for and no checklist can do: grep the old name across the project. Every occurrence, every platform, counted per folder, before anything changes.
- Your agent fixes every occurrence with judgment. Point Claude, Codex, Cursor, or Copilot at the project and give it the brief. We renamed Atlas to Meridian. Fix every unintentional mention in every folder: match the casing, repair the grammar around each swap, leave the two "formerly Atlas" lines in the help center, and list every slug, handle, or URL containing the old name instead of changing it. The agent catches what a literal replace cannot. "An Atlas account" becomes "a Meridian account", not "an Meridian account". "Atlas Pro" gets treated as the compound it is. A sentence that read fine before the swap gets read again after.
- You review one queue and publish everywhere. Every change is a word-level diff next to the original, grouped by source. Grep once more before approving: the count should equal your deliberate mentions, exactly. Approve, and Scratch writes each platform back through its own API. Because nothing ships until you approve, the rename lands everywhere at once instead of leaking out platform by platform over a month. The agent did 99% of the work. The 1% you keep is the call that it actually ships, made once, over one queue, instead of 5 times over 5 admin panels. If a published item lands wrong, revert it, per item.
Where the old name hides
The customer screenshots come from the fields no admin search covers. The grep covers them:
- Shopify SEO titles and meta descriptions, written at launch and never opened since.
- Webflow page metadata, Open Graph text, and asset alt text.
- Intercom article titles and bodies, and the collection names above them.
- HubSpot deal names, company descriptions, and the notes your sales team wrote last quarter.
- Brevo email template subjects and bodies, including templates that still send.
- Slugs and handles everywhere, which get a flag and a decision, not a silent swap.
Questions people ask
Is there one search that covers all of our platforms at once?
No. None of your platforms' own APIs can see the others. Enterprise search products do index a whole stack, but they are built to find, not to edit: even the ones with agent actions are not built to rewrite your store, help center, and CRM in place, and cross-platform find-and-replace is not something any of the mainstream ones offer. Files are the way through. Once each platform is a folder in one project, grep is the cross-platform search, and the same files take the fix.
Why not just run find and replace inside each platform?
You can. Where the tool exists, a literal swap inside one platform is quick. The rename is bigger than that. It is 5 platforms, each with its own tool or none, each with its own matching rules, each reviewed separately. The misses cluster exactly where those tools do not look: meta descriptions, alt text, deal notes, template subjects. One platform done well still leaves 4 platforms of screenshots.
Will a blind find and replace break grammar, casing, or compound names?
It can. The failure modes are well documented: replacing inside longer words, flattening the casing the sentence had, swapping half of a compound product name. The quieter one is grammar: rename Atlas to Meridian and every "an Atlas account" silently becomes "an Meridian account". This is why the fix needs an agent reading sentences, not a regex matching strings. Each repaired sentence then shows up in the diff for you to confirm.
Does renaming a Shopify product change its URL?
No. The title and the URL handle are separate fields, so after the title changes, the old name lives on in the address of the page. The same is true of Webflow slugs. Changing them breaks existing links unless you set redirects in the platform first, which makes every slug and handle a decision rather than a swap. The brief tells the agent to list them, not flip them. You choose which ones change, and you set the redirects where the platform manages them.
Should every mention of the old name disappear?
No. Customers keep searching the old name for months, and a deliberate "formerly known as Atlas" in the help center and the announcement helps them land. The target is zero unintentional mentions, not zero mentions. Tell the agent which lines are deliberate, and the final grep count should match that list exactly. When it does, you are done, and you can prove it.
Can I undo one platform's changes if they land wrong?
Yes. Approval is per item, not per run, and so is the undo. Reject a diff before publish and it never leaves your laptop. Revert an item after publish and the original goes back through the same connector. The approved changes on every other platform stay where they are.
What about the platforms Scratch does not connect?
They stay on the checklist. Social profiles, review sites, app marketplaces, and press coverage are not yours to grep, and the old name persists there no matter how clean your own stack is. Inside your stack, Scratch covers the platforms it connects. If your rename runs through one it does not, tell Curtis. Connector requests are the cheap kind.
Do I need a developer to run this?
No. Grep is a word on this page, not a skill you need. The agent runs the searches and writes any script the job calls for, like an index of every occurrence by folder. Your part is plain English: the brief, the list of deliberate mentions, and the afternoon of reading diffs.
See it on your own stack
The audit alone is worth the install: connect 2 sources, search the old name, and you have the occurrence count your checklist has been guessing at. See it run on your platforms →, or download Scratch free and pull your help center and store this afternoon. Scratch is free to try, and the AI is whichever agent you already pay for.