The short answer
Connect DuckDive to your agent from an admin seat with actions allowed, then paste your site list with each client's name. The agent adds every site under its client, links each Google source that has exactly one unused match, and returns a short list of unclear or conflicting sources. You decide those. Nothing moves between websites without your yes.
- Fix Google access before the session, not during it. The agent can only link what your login already reaches.
- Make it report three buckets: linked, unclear, and claimed by another website.
- Treat "move it" as a decision you type, never one the agent infers.
The run sheet, before you start
Bulk onboarding stalls in the same few places: a client that does not exist yet, a login that cannot see a property, two properties that both look right, and a property another website already owns. The run sheet gives each one an owner, so the agent knows when to keep going and when to stop. The middle column is the DuckDive tool doing the work; the right column is your part.
| Stage | Agent's tool | Your call |
|---|---|---|
| 1. Check the seat | get_setup_status | Admin seat, actions allowed, Google connected |
| 2. Dry run the list | list_available_sources | Fix Google access gaps it reports |
| 3. Add each site | add_website | Name the client for every row |
| 4. Link clear matches | connect_website_sources | None, it links without asking |
| 5. Settle unclear ones | list_available_sources | Pick a candidate, or skip |
| 6. Settle conflicts | connect_website_sources | Say yes to a move, or leave it |
| 7. Watch imports | get_sync_status | Chase anything that fails |
Stages follow the call order in DuckDive's agent setup guide (read November 2026).
Stages 4 and 7 run unattended. Stages 5 and 6 are where a careless batch does damage, so most of this guide lives there. Setting up one site, not forty? Skip this guide: agent self-setup takes a lone website from sign-in to first answer.
Preparing your site list and Google access
The list is the input that matters most. One row per website, three fields: the client name exactly as it appears in DuckDive, the site address, and the display name you want. Skip tracking IDs; the agent matches against what your Google login reaches.
Clean the list before you paste it
- One address per website. Write the root domain the client markets, not a landing page path or a UTM link.
- Client names spelled once. "Bramblecote Dental" and "Bramblecote Dental Group" read as two clients.
- Subdomains on their own row. Only if you want a client's shop or careers subdomain tracked apart.
Get Google access sorted first
DuckDive offers only the GA4 properties and Search Console sites your connected Google logins can see, so every site your login cannot open turns up with nothing to link. Chasing thirty grants is its own project: the guide to getting Google access for every site you report on covers which role to ask for, who can grant it and how to word the request.
Connect with actions allowed
-
Add DuckDive to your agent
If it is not there already, register
https://api.duckdive.ai/mcpas a remote MCP server in whichever app will run the batch (no API key). Per-app steps live on the Claude connector page; labels for that menu vary by app (Connectors, Apps or MCP). -
Sign in with the Google account on the most client properties
Agencies often hold client access across two or three Google accounts: an old shared inbox, a strategist's own address, a login inherited from a previous team. Count which one appears on the most properties and use it for Continue with Google. Rows only another account can open will surface in the dry run's "nothing reachable" pile, which is where you want to find them.
-
Decide who drives the session
A batch runs under one person's DuckDive role and one Google login, and the two have to belong to the same person. An admin with narrow Google access adds every row and links almost nothing; a colleague with wide access but a read-only role adds nothing at all. Pick the person who scores well on both, and let them make the connection themselves; the approval screen they will see is covered in agent self-setup.
-
Run the dry run
Before anything is created, have the agent map your list against what it can see. You get the access gaps while there is still time to chase them.
Ask your agentDry run, change nothing. For each row on my list, find the matching client in our DuckDive account and any GA4 property or Search Console site my Google access reaches. Give me three lists: ready, ambiguous, and nothing reachable.
Tools it calls
get_setup_statuslist_available_sourceslist_websites
Adding websites under the right client
In DuckDive's agency setup a website never floats free: it belongs to exactly one client. add_website takes the site address, an optional display name and a client_id. Leave the client out on a multi-client account and the tool lists your clients instead, so the agent has to ask. On a 32-row list that is 32 interruptions, unless the client is on every row. This is the core request of the session:
Ask your agentHere is a list of 32 client sites with their client names. Add each one to DuckDive under the right client, link every source in ready_to_link, and give me a list of anything unclear.
Tools it calls get_setup_status add_website list_available_sources connect_website_sources
Add two guard rails: check for sites a colleague already added last quarter, and hold any row whose client has no match rather than filing it under the nearest-sounding one.
Ask your agentBefore adding the batch, search our DuckDive account for every domain on my list and tell me which ones already exist and under which client. Skip those. If a client name on my list matches nothing in DuckDive, hold that row and tell me.
Tools it calls search list_websites get_setup_status
Adding a website imports nothing on its own; data flows once a source is linked. The DuckDive page for agencies covers how clients and roles are organised.
Letting the agent link the clear matches
For each added site, list_available_sources returns the Google properties and ads accounts your logins reach, suggested matches first, plus ready_to_link: one clear match per source type that no other website uses. DuckDive's agent guide says to link those without asking and report them. That rule is why a 32-site list does not become 64 questions.
ready_to_link- One clear, unused match per source for a given website. The agent links these without asking.
- Unclear source
- Several candidates fit. The agent asks once, as a numbered list; you can skip it.
- Already linked
- Each GA4 property or Search Console site sits on only one DuckDive website, and the listing names which.
linked_to_inaccessible_website- Held by a website your role cannot open, so no move can be approved from your session. The agent logs it in its own column and leaves it; the snags section below covers what to do when that column gets long.
move_from_other_website- Lets
connect_website_sourcestake a source from another website, only after you agree. replace_existing- Swaps the source this website already uses for another, also only after you agree.
Resolving unclear and already-linked sources
These two piles look alike in a report and are not. An unclear source is a choice between candidates nobody uses; a wrong pick is cheap to fix. An already-linked source is one somebody else uses; a wrong move strips data off a website a colleague relies on.
Why matches come back unclear
On an inherited roster, Search Console produces most of this pile. Each client verified its site whenever someone needed it, so one account carries a Domain property from a recent rebuild, another an http prefix from 2017, a third both. Google treats these as different property types with different coverage, per its help page on adding a Search Console property (as of November 2026), and the API names them differently: sc-domain: in front of a Domain property, a full URL for a prefix, per the Search Console API sites reference. Those prefixes let you sort the agent's list at a glance.
GA4 adds a different kind of duplicate: the spare property. Staging copies, the property a previous agency built and abandoned, and one property feeding a brand site plus its microsites all turn up. Google allows up to 50 data streams in one GA4 property, according to its Analytics hierarchy page (as of November 2026), so a single property serving three of a client's domains is normal, and it can show up as a candidate for each of those domains.
Write the tie-break rules once, not forty times
Answering unclear matches one by one is where a long session loses an afternoon. Instead, give the agent your house rules up front and have it apply them as proposals. You still approve, but you approve a finished list in one reply instead of nine separate questions.
Ask your agentFor every unclear source in this batch, propose a pick using my rules, then show all the proposals as one numbered list and wait. Search Console: prefer the sc-domain property, unless that client has a subdomain with its own DuckDive website. Between http and https prefixes, take https. GA4: prefer the property with a stream on the live domain. Anything named staging, test, dev or old: mark it "ask client" instead of picking.
Tools it calls list_available_sources connect_website_sources
Sources already linked elsewhere
In a batch, a conflict should be a row in a table, not a pause in the run. On a thirty-site list, you do not want the agent waiting on you about site seven while sites eight to thirty sit untouched. Tell it to log each refusal (the new site, the source it wanted, and the DuckDive website that holds that source today), carry on with the list, and hand you every conflict together at the end.
Ask your agentDon't stop on conflicts. Whenever a source you wanted is held by another DuckDive website, note the new site, the source and the website holding it, then keep going. At the end, show me every conflict in one table, numbered, and change none of them.
Tools it calls list_available_sources connect_website_sources
You then work down that table one numbered row at a time: move, swap, or leave. Only the approved rows get rerun, with move_from_other_website for a source another website holds or replace_existing where the new site already carries a different one. Your opening "link everything" never stands in for those answers; it reached only ready_to_link, and that bucket never contains a source someone else is using.
Moves are per source
Consent one source at a time, by name. "Yes to all conflicts" is how a live site loses its Search Console data to an old test website.
Worked example: onboarding thirty-two client sites in one session
Copse & Co. Agency is fictional. Its operations lead has an Agency Admin seat, a Google login her clients have added, and 32 sites across nine clients. She pastes the list with the core prompt.
Here is a list of 32 client sites with their client names. Add each one to DuckDive under the right client, link every source in ready_to_link, and give me a list of anything unclear.
get_setup_statusdriver is an Agency Admin with write access; Google linked; nine clients expected, eight foundadd_website31 calls, each with the client from the list; the Fernhill Vets row heldlist_available_sourcesonce per added site, reading ready_to_link and the unclear and in-use itemsconnect_website_sources30 calls, linking only clear, unused matches
"Done with the batch. 31 of 32 sites are in DuckDive. Here is where each one stands:"
| Outcome | Sites |
|---|---|
| GA4 and Search Console linked, importing | 24 |
| One source linked, one unclear | 4 |
| A source linked to another website | 2 |
| Added, nothing reachable to link | 1 |
| Not added, client missing | 1 |
"The unclear ones, with my pick in brackets:"
| Site | Why unclear |
|---|---|
| marrowbykitchens.example | Domain and https prefix both fit (Domain) |
| pellicottlaw.example | Live and "staging" properties (live) |
| haskettfencing.example | http and https prefixes (https) |
| sorrelbyspa.example | Two properties, one from 2021 (ask client) |
"Conflicts: Wyndham Glass's Search Console Domain property is linked to the website Wyndham Glass Commercial, and Greywick Storage's GA4 property is linked to Greywick Storage (old). I have not moved either. Quenby Tutors returned no GA4 or Search Console your login can reach."
Read the shape of that answer, not the totals. The agent did every unambiguous thing itself, stopped exactly where a wrong guess costs something, and invented no client. What is left: four numbered replies, two conflict decisions, one access request, one client to create. She replies "1, 2, 3, skip Sorrelby for now", and three more sources link.
Your agent clears the obvious matches. You keep the calls that need judgment.
Once your list is in, ask which client sites gained or lost organic clicks this month, for every client on the roster.
Free for 2 sites · No credit card
Watching the imports finish
Each linked source kicks off its own history import, so a batch does not finish at one moment. It lands in waves, and the clients with the deepest histories arrive last. Rather than opening thirty dashboards, have the agent sweep the roster with get_sync_status, which gives every source a state and the date of its newest data, and sort the result by client.
Ask your agentAcross the websites added today, which sources are still importing and which have failed? Group the failures by client, so I can see if one client's access is the cause.
Tools it calls list_websites get_sync_status
Ask soon after the batch and again later. Still importing late usually means a long history. Failing deserves a look, often at Google access removed since the dry run.
| Status | Sources |
|---|---|
| Ready | 53 |
| Importing | 3 |
| Failing | 1 |
Do not reach for a refresh. refresh_website_data After the last source turns ready, the batch is over and the portfolio operators' weekly routine takes over.
Worked example: one property claimed by two websites
Back to a conflict from the batch. Wyndham Glass, a fictional Copse client, has its main site and a commercial glazing subdomain. Last year a colleague added the subdomain as its own DuckDive website and linked it to the client's Search Console Domain property, which covers every subdomain. Now the main site needs that property, and it can belong to one website only.
The Search Console site for Wyndham Glass is already linked to another website. Tell me which one, and do not move it until I say so.
searchfinds two DuckDive websites matching "wyndham"list_available_sourcesfor the main site: the Domain property is in use elsewhere; an unused https prefix for the subdomain existsget_websitethe subdomain website's linked sourcesget_sync_statusconfirms the subdomain website's Search Console import has finished
"It is linked to Wyndham Glass Commercial, the website for commercial.wyndhamglass.example, added last year. I have not touched it. Options:"
| Option | Main site gets |
|---|---|
| 1. Move the Domain property | sc-domain:wyndhamglass.example |
| 2. Leave it | No Search Console |
| 3. Use a prefix for main | A new https prefix |
"Option 1 matches how the sites are split, since the subdomain has its own prefix property. Say 'move it' and I will run it."
Reply: Move it, option 1.
"Moved. I called connect_website_sources for the main site with move_from_other_website set, then linked the commercial site to its https prefix property. Both are importing."
Two details make this safe: the agent stopped when told, as DuckDive's consent rule requires, and it found the subdomain a new property before moving anything. Note the trade-off: the main site's Search Console totals now include the subdomain's clicks, because a Domain property counts every subdomain. Agree that with the client.
Groups folding acquired brands into one domain hit the same pattern; the guide for holding companies consolidating brand sites covers what to baseline first.
When a long batch breaks partway
Everything in this section is a problem of size. If one site refuses to add, sends you to an admin or asks you to click a Google link, that is a setup state rather than a batch fault, and the self-setup page's stall table decodes it from the status fields. What follows only happens once the list is long.
The chat ran out of room at row 23
Long conversations hit limits set by your AI app, not by DuckDive, and every row spends several tool calls: an add, a source listing, one or two links. Feed rows in groups of about ten and ask for a running tally after each group. If a chat does die, resist pasting the whole sheet into a fresh one. Rows already added stay in DuckDive, so the first job in the new chat is a diff against what exists.
Ask your agentA previous chat stopped partway through onboarding this sheet. Page through every website in our DuckDive account first. Then split my rows into three groups: already added with sources linked, already added with nothing linked, and not added. Resume with the second and third groups only, ten rows at a time, and give me a tally after each ten.
Tools it calls list_websites get_website add_website list_available_sources connect_website_sources
The account already holds more than 100 websites
A single list_websites call stops at 100 websites, or 50 if the agent does not ask for more; anything past that needs another call with an offset. An agent that skips the paging treats page one as the whole account, misses sites a colleague added, and tries to create them again. Make it say how many websites it counted in total, and check that figure against your own records before the first add.
Nine rows held because the client "does not exist"
A cluster of held rows nearly always traces back to the sheet, not to DuckDive: "Ottersley Dental" on your list, "Ottersley Dental Group" in the account. Have the agent print its client names next to the distinct names in your client column, fix the column once, and rerun only the held rows.
The conflict log is longer than expected
Twenty conflicts on a forty-site list usually share a handful of holders: websites created during a trial, a sandbox someone built to test reports, the "old" copy left behind by a migration. Ask the agent to regroup the log by holding website. Each source still gets its own yes or no, but a group whose holder imports nothing current is a quick sitting.
Rows flagged linked_to_inaccessible_website regroup the same way, and grouping matters more there, because each holder answers to an admin you have to contact. One message per holder, listing every source your new sites need from it, beats a dozen separate requests.
| Holding website | Can she approve moves? |
|---|---|
| Copse sandbox (spring trial) | Yes, one by one |
| Ottersley Dental (old domain) | Yes, after asking the client |
| A website under Wrenfield Bakeries | No, outside her role; one note to that client's admin |
Failures pile up under one client
When several failing sources share a client, look at that client's Google side before anything else: a user removed during a staff change, or a property owner who left. One request to that client clears the group. A lone failure elsewhere on the roster deserves its own look.
Ask your agentAudit today's batch. Grouped by client, list every website we added with which of GA4 and Search Console are linked, which are still empty, and why each empty one is empty.
Tools it calls list_websites get_website list_available_sources
Working in a terminal? Claude Code can read the list from a CSV in your project folder; the Claude Code setup page has the connect command.
Who clears the leftovers
A good batch ends with a short remainder, and none of it is the agent's to finish. Assign each pile an owner before you close the chat, or the same rows will still be empty at next month's report.
| Pile | Owner | Where it gets done |
|---|---|---|
| Nothing reachable | The client, or whoever manages their Google accounts | In Google; the access request guide has the wording |
| Held for a missing client | You or another agency admin | The DuckDive app, then rerun the held rows |
| Added by mistake | An admin | The DuckDive app, since the connector cannot delete |
| Yext and Local Falcon | Whoever holds those logins | The integrations link in your setup status |
| No tracking on the site at all | Whoever builds the client's site | Outside DuckDive: a tag has to exist before there is anything to link |
And sometimes a batch is the wrong tool. Under about five sites, adding them through the in-app welcome flow takes less time than preparing a sheet. If the real need is branded PDF reports for clients rather than answers to questions, a reporting dashboard product is the closer fit.
Before you hand over the list
Can my AI agent add many websites to DuckDive at once?
Yes, though not in a single call. The agent loops over your rows: add_website, then list_available_sources and connect_website_sources for that site, then the next row. Feed it about ten rows at a time with a running tally so a long list survives your AI app's own chat limits.
Will the agent move a GA4 property or Search Console site off another website by itself?
No. A source that another DuckDive website already holds goes into the conflict table the agent gives you at the end of the batch, next to the name of the website holding it. Nothing on that table changes until you approve its row by name; a blanket instruction to link everything, given at the start of the session, is not an approval.
How do I make sure each site lands under the right client?
Put the client name on every row, spelled as it appears in DuckDive. On a multi-client account, add_website needs a client_id; without one, the tool lists your clients and the agent has to ask. Create any missing client in the app first.
Why did some sources come back unclear?
More than one candidate fits: a Search Console Domain property beside a URL-prefix one, separate http and https prefixes, or a staging GA4 property next to the live one. Give the agent tie-break rules at the start of the batch and it returns one list of proposed picks for you to approve together.
What happens to rows my Google login cannot reach?
They come back with nothing to link. Run a dry run before anything is created and you get that list while there is still time to ask each client for access; rows added anyway sit empty until a source is linked.
What if the chat stops halfway through the list?
Start a new chat and have the agent compare your sheet with every website in the account, paging past the first 50 or 100, before it adds anything. Rows already added stay added, so it resumes from the rows that are missing or still have no linked sources.
Sources
- Set up DuckDive with your AI assistant (agents.md). DuckDive AI (opened November 16, 2026).
- Add a website or platform property to Search Console. Search Console Help (opened November 16, 2026).
- Sites (Search Console API reference). Google for Developers (opened November 16, 2026).
- Google Analytics hierarchy. Google Analytics Help (opened November 16, 2026).
- DuckDive for Claude, ChatGPT and other AI apps. DuckDive AI (opened November 16, 2026).
Last reviewed . Copse & Co. Agency, Wyndham Glass and every other business and site in EXAMPLE DATA blocks are fictional.
