For in-house marketing teams

Answer leadership's marketing questions before the meeting, straight from your agent

The CEO asks in chat why leads are down, and finance will pick apart whatever you send back. This is how a head of marketing or marketing ops replies with sourced GA4 and Search Console numbers for the main site, docs and blog.

Continue with Google

Free for 2 sites · No credit card

Or have your agent set it up

Set up DuckDive for my website. Follow https://analytics.duckdive.ai/agents.md

Updated 16 min read

The short answer

Connect DuckDive to the AI agent your team already uses and add the main site, docs and blog as three separate websites. When an executive asks a marketing question, have the agent reply in four parts: the claim, the number with its counts, the source and exact window, and the next check. Say so plainly when the data cannot answer.

  • Write the window as dates, because the tools use rolling ranges such as the last 90 days, never calendar quarters
  • Split brand searches out before anyone calls search a demand engine
  • Send pipeline and revenue questions to the CRM, with one sentence on why

Five questions leadership asks, and the path to each answer

Executive questions arrive half-formed in a chat thread, an hour before a meeting. Each hides a choice of website, metric and window, and finance will ask which one you made. Here are five common ones, the DuckDive tools that answer each, and the trap inside it.

Behind the answers: DuckDive is a remote MCP server your agent signs in to with Google. It imports and keeps each website's GA4 and Search Console history (plus Google Ads, if you run it), so your agent reads stored numbers while you skip the export.

"Why are leads down?" (the CEO, in chat)

They mean demo requests, this week, on the main site. The path: get_ga4_overview to size the drop by day, get_ga4_channels for the last 7 and 14 days (the earlier week is the difference), then get_ga4_landing_pages. The trap is small numbers: a few dozen weekly conversions move a lot on chance, so carry counts, not just a percentage.

Ask your agent
The CEO wants to know why demo requests look soft. On the main site, set this week against last week by channel, then show the landing pages behind the biggest gap. Tell me if the counts are too small to call it.

Tools it calls get_ga4_overview get_ga4_channels get_ga4_landing_pages

"How much of our demand starts in search?" (the board deck)

This goes on a slide and stays there for a year. The path is get_ga4_channels for the shares, then get_search_console_keywords to see how much search traffic is people typing your brand name. See the first worked example.

"Is the docs site worth the writer we hired?" (the product lead)

The real question is whether people find the docs, so start in Search Console rather than GA4. The path is get_search_console_summary on docs and on the main site over 6 months. The trap is a Search Console property covering both, which folds docs clicks into the main site's totals; keep them apart.

"Is the new pricing page pulling its weight?" (the revenue lead)

The path is get_ga4_landing_pages in two windows and get_search_console_pages for Google clicks. The trap: a landing page is only the first page of a visit. The second worked example handles it.

"Should we keep paying for the blog?" (the CFO)

The path is get_ga4_landing_pages on the blog website and get_search_console_keywords for its searches. The trap: if readers sign up later on the main site and the blog has its own GA4 property, the blog shows almost no conversions while still doing its job.

Ask your agent
Finance is asking whether the blog earns its budget. For the blog website over the last 6 months, list posts that started converting visits and the top queries that brought readers in, then say what this data can't show.

Tools it calls list_websites get_ga4_landing_pages get_search_console_keywords

An answer format that survives the finance team

Finance rarely distrusts marketing numbers because they are wrong. It distrusts them because they arrive without their working. "Organic is up" invites three questions: up from what, over which dates, measured where. Answer them first and the thread ends with you. Use four labeled lines, always in this order.

Claim
One sentence an executive can repeat, naming the website, because "the site" means three things once you have docs and a blog.
Number
The figure, with the two counts that make any percentage.
Source
System, DuckDive tool, the metric's name in that system, and the window as dates. "GA4, sessions by medium, the 90 days ending November 15" beats "Google Analytics".
Next check
The one test that could prove the claim wrong, and who runs it. It tells finance you looked for the hole yourself.

Two vocabulary points stop most arguments. GA4 calls the actions that matter to your business key events and by default counts each time one fires. DuckDive's tools call them conversions, so write "conversions (GA4 key events)". And Google says Google Ads conversions don't appear in standard GA4 reports, which is why the two "conversions" numbers rarely match. Name the one you used.

Before a reply goes to leadership

  • Dates, not labels. "Last quarter" becomes the start and end days the agent used.
  • One website per number. Never add main site, docs and blog without saying so.
  • In and out, in a clause. Bing sits in GA4's organic but not in Search Console; offline deals sit in neither.
  • A check that could fail. If the next check cannot change your mind, it is not a check.

Make it a standing instruction, pasted once into your agent's custom instructions or project notes.

Ask your agent
Whenever I forward a question from leadership, answer in four labeled lines. Claim: one sentence naming the website. Number: the figure with the counts behind it. Source: system, DuckDive tool, metric name and the exact dates of the window. Next check: what could prove this wrong. If the data can't answer the question, say that in the Claim line and stop.

Tools it calls list_websites get_sync_status

The format makes the work checkable; it does not check the agent. For that, see checking your AI agent's marketing numbers.

Worked example: the board asks how much demand starts in search

Saltgrass Software is a made-up company selling shift scheduling to field crews. The board wrote "last quarter", but the agent's tools take rolling windows (anything from one week to one year, always ending today), so the honest reply uses the last 90 days and says so.

EXAMPLE DATAFictional business: Saltgrass Software

The board wants to know what share of last quarter's sessions and conversions started from organic search. Answer as claim, number, source and next check.

  1. list_websitesfinds three websites: Saltgrass main site, docs and blog, and picks the main site
  2. get_ga4_channelsmain site sessions and conversions by medium, last 90 days
  3. get_search_console_summarymain site Google Search clicks for the same window
  4. get_search_console_keywordsthe top 100 queries by clicks, to separate brand searches from the rest

Claim: On the Saltgrass main site, organic search started about two in five visits and one in three conversions, and much of it was people already searching for Saltgrass by name.

Number: 43% of sessions (21,300 of 49,000) and 34% of conversions (286 of 840). Of the 18,900 Google clicks in the window, at least 6,800 (36%) came from searches containing "saltgrass".

MediumSessionsConversionsShare
organic21,30028634%
(none), direct14,90024129%
cpc8,20019824%
referral3,100617%
email1,500546%
Total49,000840100%

Source: GA4 via DuckDive, sessions and conversions (key events) by medium, main site, the 90 days ending November 15, 2026, not the calendar quarter. Brand split from Search Console's top 100 queries; 7,560 clicks sit outside that list or are hidden, so 36% is a floor.

Next check: For July 1 to September 30 exactly, pull GA4 itself. Rerun each quarter: the non-brand share shows whether search finds new buyers.

GA4 Search Console

Three details make this hold up. First, the systems count different things. Google's default channel group files traffic as Organic Search when the source is on its list of search sites or the medium is exactly "organic", so Bing counts too. Search Console reports only how your site performs in Google Search results, in clicks. Expect sessions above clicks.

Second, the brand split is a floor. Google hides anonymized queries, meaning any query that fewer than a few dozen people searched across two or three months, so query rows never sum to the total. Say "at least".

Third, day boundaries. Search Console labels days by local time in California, while GA4 uses the reporting time zone set on the property. Over 90 days that barely matters. For one bad Monday it can.

Put your own search share in front of the board

Link your main site and ask the board question above; the brand split comes back with it.

Continue with Google

Free for 2 sites · No credit card

Worked example: is the new pricing page pulling its weight?

Six weeks ago Saltgrass shipped a redesigned pricing page, and the revenue lead wants a verdict before the next change. The tools have a 90-day window and a 6-month window but nothing in between, so the agent derives the earlier window by subtraction and says it is approximate.

EXAMPLE DATAFictional business: Saltgrass Software

Is /pricing converting better in the last 90 days than in the 90 before, and did its search clicks change?

  1. get_ga4_landing_pagesmain site, last 90 days, the /pricing row
  2. get_ga4_landing_pagesmain site, last 6 months, so the earlier window is the difference
  3. get_search_console_pagesGoogle clicks to the pricing URL, last 90 days and last 6 months

Claim: Yes. Visits that start on /pricing convert more often than before, and the gain is not coming from Google Search.

Number: 5.0% of sessions landing on /pricing converted (162 of 3,240), against 3.6% (94 of 2,610) in the earlier window. Google clicks to the page slipped from 1,240 to 1,180.

/pricingEarlier windowLast 90 daysChange
Sessions landing here2,6103,240+24%
Conversions94162+72%
Conversion rate3.6%5.0%+1.4 pts
Google clicks1,2401,180-5%

Source: GA4 landing pages (conversions are key events) and Search Console pages, main site. The earlier window is the 6-month figure with the latest 90 days taken out, which runs a few days past 90. Only sessions whose first page was /pricing count.

Next check: The last 90 days still hold about seven weeks of the old page, so rerun on the last 30. Search clicks fell while landings rose: ask the paid team whether a campaign now points here.

GA4 Search Console

People skip the caveat in the Source line. In Google's definition, a landing page is the path of whichever page a session viewed first. Buyers who reach pricing from the homepage never appear in this row, so this judges pricing as a front door, not as the third stop in a visit. If the question becomes "which pages get found but don't produce demos", use the method in finding pages that rank but don't convert.

Main site, docs and blog as separate websites

Buyers read the main site, customers read the docs, and the blog sits between. Add them up and a docs surge can hide a main-site slump on a board slide. DuckDive can hold the three apart: point the docs website at the docs analytics and the main website at its own, and whatever is linked to one stays off the other. Your Google setup decides whether that split is clean.

Search Console: watch the domain property

A Search Console Domain property includes all subdomains, so it counts www and docs together. A URL-prefix property covers only URLs under its prefix. Verify one for the docs host and one for www, and link each to its own DuckDive website.

GA4: one property or three

If docs or the blog report into the main site's GA4 property, DuckDive cannot split them, and since a landing page is a path without the host name, same-named paths merge into one row. Link the shared property to the main website and give docs its Search Console site only. That still answers the usual docs question, which is about search.

Ask your agent
Is the docs site's search traffic growing faster than the main site's over the last 6 months?

Tools it calls list_websites get_search_console_summary get_search_console_keywords

EXAMPLE DATASaltgrass Software: Google Search clicks per 30-day block, main site and docs, last 6 months
30 days endingMain siteDocs
June 186,0202,150
July 186,1102,330
August 175,9802,540
September 166,2402,710
October 166,3002,960
November 156,3803,180
First block to last+6%+48%

Here docs climb every block while the main site barely moves. Before crediting the docs writer with new demand, check the docs site's top queries. If most name the product plus a feature, the growth is customers looking for help: good for retention, weak evidence of new buyers.

When the honest answer is "this data cannot tell us"

Answer a pipeline question with a traffic number and finance stops trusting you. GA4 and Search Console know visits, clicks and key events, not which demo became a deal. DuckDive draws only on GA4, Search Console, Google Ads, Local Services Ads, Yext and Local Falcon; your CRM, billing and social ad accounts are not among them.

What to say when leadership asks something this data cannot settle
QuestionDuckDive dataLook in
What pipeline did the website create?Partly, demo requests by channel and pageThe CRM, using its lead source field
What revenue came from search?No, no revenue or deal dataCRM closed deals joined to first touch
What does a customer cost by channel?Partly, Google Ads spend and cost per conversionFinance's spend ledger plus closed deals
Did the LinkedIn campaign work?No, not a sourceLinkedIn's own campaign reports
Same quarter two years ago?No, agent ranges stop at 12 monthsGA4's own reports for older dates

"This data cannot tell us" is not a dead end when you add what it can tell you.

Ask your agent
The CFO asked how much pipeline the website created over the last 90 days. Tell me which part of that our DuckDive data can support and which part it can't, then draft a two-sentence reply that gives the demo request count by channel and says the pipeline figure has to come from the CRM.

Tools it calls get_ga4_channels get_google_ads_summary

Two more limits. With consent mode, GA4 may use behavioral modeling to estimate users who decline analytics cookies, so some sessions are estimates. And numbers are only as fresh as DuckDive's last import, shown by get_sync_status.

Teams that should skip DuckDive

  • Leadership wants a dashboard to open themselves. GA4's reports or a Looker Studio board suit that better.
  • Most executive questions are about LinkedIn, Meta or pipeline, none of which DuckDive reads.
  • GA4 already flows into a warehouse with an analyst on it, who can run custom ranges and joins these tools cannot.

Who on the team gets admin rights, who reads

Admins set things up: add websites, link GA4 and Search Console, queue a data refresh. Team members only read, never change; which websites each one sees follows their DuckDive access. An agent works with its person's role, and takes setup actions only if that person is an admin who kept Also allow actions ticked when connecting.

A starting split of DuckDive roles for an in-house team
PersonRoleConnection
Marketing ops leadAdmin, owns setupAllows actions at setup; reconnects read-only afterwards
Head of marketingAdmin, as the backupRead-only for daily questions
Content, SEO and demand genTeam member, readsRead-only
Finance analystTeam member, readsRead-only, to rerun your prompts

Don't skip the finance row. When the analyst pastes your prompt into their own agent and gets the same figure, the fight over whose number is right never starts.

Keep admins to two, since each can relink sources. The connector cannot delete data. DuckDive logs which tool ran, for which website and when, but never receives the conversation; once results reach Claude, ChatGPT or Cursor, that vendor's terms decide how they are kept. For a security review, read whether it's safe to connect AI to Google Analytics.

Rolling it out to the team

The marketing ops lead handles the first four steps alone; only the last two involve the rest of the team.

  1. Pick the Google login that sees all three sites

    Use a Google account that can see every GA4 property and Search Console site involved.

  2. Sign up, or let the agent do it

    Choose Continue with Google above, or paste the setup prompt. The agent registers DuckDive and asks you to click Continue with Google once; it never sees your password.

    Paste this into your agent
    Set up DuckDive for my website. Follow https://analytics.duckdive.ai/agents.md
  3. Add the main site, docs and blog one by one

    The agent calls get_setup_status, which suggests websites from your Google account; tell it all three. It links each site's clearly matching GA4 property and Search Console site, and asks once about anything unclear, such as a Domain property that fits two sites. Link Google Ads to the main site too, if it runs ads.

  4. Wait for the history import

    Each source starts importing its history the moment it is linked, and sites with years of data take longer than new ones. Have the agent confirm with get_sync_status that each source is ready before anyone quotes a number.

  5. Add teammates as read-only members

    Invite the people in the roles table. Every teammate links DuckDive to their own assistant, with Also allow actions left unticked.

  6. Share the format and the prompts

    Put the four-line instruction and these prompts in the team wiki, so the same question asked twice gets the same number.

The Claude setup page has a link that fills in the connector for you. Field definitions for sessions, key events and channels are on the GA4 integration page. A founder's personal morning brief is a different setup, covered on the page for founders running an AI chief of staff.

What marketing leads ask before rolling it out

Can the agent pull the exact calendar quarter the board means?

Not today. Every data tool counts back from today, in windows as short as a week or as long as a year. Use the last 90 days and write its start and end dates in the Source line. For an exact calendar quarter, pull it from GA4's own reports and say so.

Why doesn't the agent's organic number match the GA4 report finance opened?

Check three things. The window: a rolling 90 days is not a calendar quarter. The grouping: check whether the agent's rows are medium values (organic, cpc) or channel groups (Organic Search), since a GA4 report grouped by channel can file the same visits differently. The metric: tool conversions are GA4 key events, not Google Ads conversions.

Do the main site, docs and blog each count as a site for pricing?

Yes, each is a separate website in DuckDive. One of the three can sit on the free plan with no credit card; the cost of adding the other two is in the plans further down. Start with the main site, then add docs and the blog when people ask about them.

Can the agent see our CRM pipeline or revenue?

No. None of DuckDive's six sources holds deals or invoices; pipeline and revenue live in your CRM and billing system. Your agent may have a separate connector for those, but DuckDive does not join them.

Which AI assistant should the team standardize on?

Pick the assistant your company already licenses, provided it accepts remote MCP connectors that sign in through OAuth. ChatGPT qualifies only on Business or Enterprise; any Claude plan, Cursor and Claude Code all qualify. Since every teammate links DuckDive to their own assistant, the content lead on Claude and the analyst in Cursor still read the same numbers.

What happens when someone leaves the team?

On their last day, have them open the AI Connectors page in DuckDive and disconnect, or delete the DuckDive connector from their AI app; either one ends that assistant's access at once. If they were the admin whose Google login linked your sources, watch get_sync_status for a few days and relink from another admin if a source stalls.

Sources

  1. About key events, Google Analytics Help, checked November 16, 2026.
  2. Conversions vs. key events in Google Analytics, Google Analytics Help, checked November 16, 2026.
  3. Default channel group, Google Analytics Help, checked November 16, 2026.
  4. Performance report (Search results): Overview and basic setup, Search Console Help, checked November 16, 2026.
  5. A deep dive into Search Console performance data filtering and limits, Google Search Central Blog, checked November 16, 2026.
  6. About Search Console data, Search Console Help, checked November 16, 2026.
  7. Resource: Property, Google Analytics Admin API, checked November 16, 2026.
  8. Analytics dimensions and metrics, Google Analytics Help, checked November 16, 2026.
  9. Add a website or platform property to Search Console, Search Console Help, checked November 16, 2026.
  10. [GA4] Behavioral modeling for consent mode, Google Analytics Help, checked November 16, 2026.

Last reviewed . Numbers in EXAMPLE DATA blocks are fictional.

Have the answer ready when the question lands

Link the main site, docs and blog once, then reply with numbers finance can trace to GA4 and Search Console.

Free for 2 sites · No credit card · Working in an agent? Let it set DuckDive up