The short answer
Ask each founder for read-only access to Google Analytics 4 and Search Console, using opt-in wording that says what you will see and how to withdraw. Once the consenting sites sit in DuckDive, your agent can rank signups, organic share and branded search across the whole cohort. Founders can revoke access in two Google settings screens.
- Consent first: one written opt-in per company, with scope, audience and an exit
- Read signups as conversions, the organic share of sessions, and searches for the company's own name
- Leave a company out of rankings until it clears a traffic floor and has two full comparison windows
Asking founders for access: the consent playbook
Your information rights probably cover a quarterly update. They do not put your email on a founder's Google Analytics property. Analytics access is a login inside the company's own Google accounts, so it needs its own yes. Run the ask the same way for every company:
- Decide the purpose first. "So we can spot who needs growth help" is a purpose founders can weigh. "Portfolio monitoring" is not. If the numbers will reach follow-on discussions, say so now.
- Ask for one shared platform-team login. Founders then see one clearly named address in their user list instead of a scatter of personal emails, and access stays with the firm when an associate leaves. Name the people behind that address in the opt-in note itself.
- Ask for the lowest read-only role in each tool. The role to request from each Google product is spelled out in our access guide, so founders are never asked for more than viewing rights.
- Ask for two tools, not four. GA4 and Search Console carry almost every seed-stage signal. Request Google Ads only from companies that want spend advice.
- Log every yes, scope change and withdrawal. Date, scope, who approved, and when access ended.
- Ask again at each new round. A founder who agreed at pre-seed may want a narrower scope later.
Withdrawal has to be as easy as the yes. In Analytics, the founder opens Admin, then Access Management, ticks your login and removes it. Google notes that deleting a user needs the Administrator role at the account level (Add, edit, and delete Analytics users, as of November 2026). In Search Console, an owner opens Settings, then Users and permissions, and chooses Remove access (Managing owners, users, and permissions, as of November 2026). Put both paths in your opt-in so nobody has to email you to leave.
| Opt-in level | Founder grants | Your agent can answer |
|---|---|---|
| Search only | Restricted user in Search Console | Partly: clicks, queries, branded share |
| Search and site | Adds read-only access to GA4 | Yes: signups, channels, landing pages |
| Paid search too | Adds read access to Google Ads | Yes: account spend and cost per conversion |
| Revenue or product data | Not part of this ask | No: DuckDive does not read these |
Based on the data DuckDive's tools return as of November 2026. Paid search shows up as totals for the whole Ads account, never per campaign.
Opt-in wording founders can say yes to
Founders read vague requests as surveillance and specific ones as help. The note below names the data, the people, the use and the exit. Edit the brackets, and delete any promise your team cannot keep.
Opt-in note to send each founderHi [first name], We'd like read-only access to [company]'s Google Analytics 4 and Search Console, so we can spot early where we could help. What we would see: daily visits and signups, traffic by channel, top landing pages, and Google search clicks and queries. What we would not see: individual visitors, revenue, or your product data. Who sees it: [names] on the platform team, through an AI assistant connected to DuckDive. The connector can't delete data, and with the read-only role you grant us, we can't change your settings. How we use it: [support conversations only / also in portfolio reviews]. Withdraw any time: remove [platform@yourfund.com] in Analytics (Admin, Access Management) and in Search Console (Settings, Users and permissions). Or reply "stop" and we'll disconnect within [n] working days. Saying no changes nothing about how we work with you.
"What we would not see" answers the fear before it is raised. "Who sees it" turns a fund into named people. The last line only works if it is true, so have the partner who runs the program sign off on it first.
Traction signals that matter at seed stage
A seed company's session count swings with every launch post. On its own it says little about pull. These readings say more:
- Signups (conversions)
- The GA4 key events the founder marked, such as account creation. Google describes a key event as one that "measures an action that's particularly important to the success of your business" (About key events). DuckDive's tools report them as conversions.
- Signup rate
- Conversions divided by sessions in the same window.
- Organic share
- Organic search sessions divided by all sessions. Low at launch, and slow to fake.
- Branded click share
- The part of Search Console clicks from searches for the company's own name.
- Sessions floor
- The minimum sessions a company needs in a window before you rank its percentage change.
Ask founders to name their signup event the same way. Google's recommended events include sign_up, described as when a user "signs up for an account on your website or app" (Recommended events, as of November 2026). Then the batch's rates line up.
Event drift
A founder who adds a waitlist or newsletter event as a second key event doubles the conversion count overnight without one new user. When a company's rate jumps, ask what changed in GA4 before you send congratulations.
For how to compare rates fairly across companies of very different sizes, with medians and floors worked out, use our guide to benchmarking conversion rates across clients.
Branded search as an early demand signal
When more people type a startup's name into Google, word is spreading. Ads can't buy that, and it often shows in Search Console well before rankings for generic terms do.
Google added a branded queries filter to Search Console in November 2025. It is AI-assisted, covers misspellings and product names, and works only for top-level properties "with a sufficient volume of queries and impressions" (Introducing the branded queries filter in Search Console). Many seed companies will not qualify yet. So count clicks from the top queries that contain the company's name.
Ask your agentFor Patchwork Robotics, what share of Search Console clicks in its top 100 queries over the last 90 days came from queries containing its own name, against the 90 days before?
Tools it calls list_websites get_search_console_keywords
The agent pulls the top 100 queries over the latest quarter and again over the half-year, then subtracts one from the other to approximate the quarter before. Queries that sit in one list but not the other make that estimate rough, so read the direction, not the decimal.
| Window | All clicks | Name clicks | Share |
|---|---|---|---|
| Prior 90 days | 58 | 49 | 84% |
| Last 90 days | 131 | 101 | 77% |
Here a falling share is good news. Name searches roughly doubled, from 49 clicks to 101, and clicks from other queries grew from 9 to 30. The totals are tiny, as they should be for a hardware company still running a waitlist, but the direction says the company is starting to be found by people who did not know it yet. Three traps to watch:
- Common-word names. "Patchwork" alone would match quilting searches. Match the full name, or the name plus the product noun.
- Hidden rare queries. Google leaves out queries not "issued by more than a few dozen users over a two-to-three month period", though their clicks still count in the totals (Search Console performance data filtering and limits). Small sites lose a large slice of rows this way.
- Launch spikes. Compare 90-day windows, not single weeks.
Worked example: a fourteen-company batch eight weeks after demo day
The agent can only find a batch by what it can search, and its website search matches a site's name or address. So when each startup goes in, its DuckDive name opens with a batch code: "S26 Tallyfern", "S26 Orbitwell". One code then returns the whole spring batch. The match also runs on addresses, so check that the count the agent reports equals your batch list before trusting the ranking.
Ask your agentFor every website whose name starts with S26, compare conversions in the last 30 days with the 30 before and rank them by percentage change, skipping any with fewer than 200 sessions.
Tools it calls list_websites get_ga4_overview
Rank every S26 site by change in conversions, last 30 days against the 30 before. Skip anyone under 200 sessions.
list_websitessearches for "S26" and gets back fourteen websites whose names carry the batch codeget_ga4_overvieweach site, 90 days of daily sessions and conversions, summed into the latest 30 days and the 30 before
Twelve companies cleared the 200-session floor. Tallyfern grew signups fastest, from 38 to 71. Read the lower half carefully: demo day falls inside the earlier window, so a drop there often means the spike rolled off, not that the company stalled.
| Company | Sessions | Signups |
|---|---|---|
| Tallyfern | 2,840 | 38 to 71 |
| Sprigloop | 1,960 | 52 to 84 |
| Orbitwell | 6,100 | 120 to 172 |
| Dovetail Ledger | 3,050 | 64 to 85 |
| Pollenware | 9,400 | 210 to 252 |
| Inkharbor | 1,120 | 45 to 51 |
| Ridgecart | 12,800 | 300 to 318 |
| Tidewright | 4,300 | 88 to 89 |
| Brackenbyte | 5,200 | 140 to 131 |
| Lumenvault | 3,900 | 96 to 74 |
| Snowmoth Labs | 870 | 33 to 24 |
| Moxa Fleet | 2,150 | 58 to 39 |
Skipped for low traffic: Patchwork Robotics (140 sessions, a hardware waitlist) and Zephyrsum (95 sessions, GA4 added three weeks ago).
Two worth a conversation. Moxa Fleet has the steepest fall above the floor; check whether its signup event changed. Lumenvault fell 23% on 3,900 sessions, so that drop is not small-number noise. Snowmoth Labs is 9 signups down; look again next month before raising it.
The prompt spans sixty days, yet the trace shows a 90-day pull. There is no way to hand the agent arbitrary start and end dates, only a short menu of preset look-backs, and the quarter is the smallest one reaching past day sixty. The daily rows then get summed into your two halves. Write the run date on the saved table: asking again next Monday slides both halves forward and demo day drifts toward the edge.
See your own batch ranked the same way
Connect one consenting company free, then ask which pages turn its demo-day visitors into signups.
Free for 2 sites · No credit card
Agenda items for the platform team's weekly sync
Data only helps if it changes who gets a call this week. Twenty minutes, same order, one person driving the agent:
- Data health, two minutes. Which sites stopped updating? Usually a broken tag after a rebuild, or a founder who withdrew access.
- Movers, five minutes. The three biggest week-on-week changes in signups, either direction, and the channel behind each.
- One company up close, eight minutes. Rotate through the batch, read its channels and landing pages, and agree one offer of help.
- Founder asks, three minutes. Match open requests to what the numbers show.
- Consent log, two minutes. New opt-ins, scope changes, withdrawals. Disconnect withdrawn sites the same week.
Ask your agentAhead of Thursday's platform sync: among the websites named S26, which three companies had the biggest change in signups this week against last week, up or down, and which channel moved most for each?
Tools it calls list_websites get_ga4_overview get_ga4_channels
Keep the full batch table inside the platform team. Ranking founders in front of each other turns support into a scoreboard, and the next opt-in gets harder.
Worked example: spotting a founder who needs help with search
Companies with Search Console linked but almost no Google traffic are easy wins for a team with an SEO advisor on call.
Ask your agentWhich portfolio sites have Search Console connected but get under 5% of their sessions from organic search?
Tools it calls list_websites get_ga4_channels
Lumenvault came back on that list. Rather than forward a chart, the platform lead asks the agent to write the founder a note.
Draft a short, friendly note to the founder of Lumenvault explaining its signup trend over the last 90 days and one page worth fixing first.
list_websitesfinds Lumenvault and confirms GA4 and Search Console are linkedget_ga4_overview90 days of daily sessions and conversions, grouped into three 30-day blocksget_ga4_channelssessions and conversions by channel for the 90 daysget_ga4_landing_pagestop landing pages with conversion rates for the 90 daysget_search_console_summaryclicks, impressions and average position for the 90 days
For you first: signups ran 81, then 96 in the demo-day block, then 74. Organic search brought 350 of 11,400 sessions (3.1%) and 6 of 251 signups. Search Console shows 18,600 impressions at an average position of 31.4: listed, but far down. One page stands out:
| Landing page | Sessions | Signups |
|---|---|---|
| / | 6,900 | 188 |
| /blog/launch-week | 1,800 | 4 |
| /pricing | 1,250 | 41 |
| /docs/quickstart | 820 | 14 |
| /integrations | 330 | 4 |
Draft note:
"Hi there, a quick read on Lumenvault's last 90 days. Signups went 81, then 96 around demo day, then 74. That dip looks like the launch bump fading, and your homepage still converts at 2.7%. One page worth fixing first: the launch-week post. It brought 1,800 visits and 4 signups. Worth checking whether it links to signup near the top. Separately, Google search sends you about 3% of visits so far. If you'd like, we can set up 30 minutes with our search advisor. No pressure either way."
The agent cannot see the post, so it suggests a check rather than a cause. Read the draft before it goes out, and if the numbers will be quoted back to you, open GA4 and confirm one figure by hand first; the walkthrough on checking an agent's marketing numbers shows where to look.
Startups with no GA4 yet
Some companies in an early batch will have no analytics at all. Help them start, in this order.
- Search Console first. Google says data collection for a property starts as soon as anyone adds it to Search Console, even before verification (Add a website property to Search Console).
- Then a GA4 property and web stream. Google's setup is account, property, web data stream, then the tag, and "data collection may take up to 30 minutes to begin" (Set up Analytics for a website and/or app). Reports start from the day the tag goes live.
- Mark one signup key event. Same event name as the rest of the batch.
- Then grant your platform login. Only after the company is happy with what it collects.
DuckDive imports whatever history exists, which for a new property is a few weeks. Keep the company out of rankings until it has two full windows.
Ask your agentZephyrsum put GA4 on its site three weeks ago. Is its data importing yet, and how many days of sessions are there so far? Tell me the date it will have 60 full days so I can add it to the batch table.
Tools it calls search get_sync_status get_ga4_overview
Founders who want these numbers for themselves can connect their own agent; see giving a founder's AI chief of staff marketing data.
Connecting a cohort, start to finish
DuckDive sits between founders' Google accounts and your team's AI agent. Connect each consenting site once; DuckDive stores its history by website, so the agent reads stored numbers instead of calling Google for every company on every question.
-
Create the account with the platform login
Use Continue with Google with the shared account founders will grant access to.
-
Send the opt-in and wait for grants
Nothing moves until founders add your login, and that is the point.
-
Hand setup to your agent
Paste the prompt below into Claude, ChatGPT (Business or Enterprise), Cursor or any client that supports remote MCP with OAuth. You click Continue with Google once; keep "Also allow actions" ticked and use an admin account. See letting your agent set DuckDive up.
Paste this into your agentSet up DuckDive for my website. Follow https://analytics.duckdive.ai/
agents.md -
Add each startup with its batch code in the name
When the agent adds a site, give it the display name to use, such as "S26 Patchwork Robotics", rather than letting it default to the domain. Pick one code format for every batch, say a season letter plus the year, and never reuse it; the page on naming sites so the agent finds them fast covers naming habits in more depth.
-
Link the sources that exist
One GA4 property and one Search Console site per startup, plus Google Ads if the founder shared it. If a founder runs two products off one analytics property, pick which DuckDive site gets it, and expect a confirmation question if that property is already attached to another site.
-
Check the import, then run the first ranking
The import usually takes a few minutes. Ask for sync status if a company looks empty.
Not ready to set up an agent? Drake, the assistant inside DuckDive, takes questions without one. Partners who want the security story first can read what a connected agent can and cannot touch in Google Analytics, including what gets logged.
Where this falls short for a platform team
Tell partners what this cannot answer before anyone builds a review on it.
- Product traction. Activation, retention, revenue and MRR live in product analytics and billing. For a software startup, the useful feeds here are usually just its website analytics, its Google search data and any paid search; the local listing and map-ranking sources mostly matter to storefront businesses.
- Signups that skip the website. If accounts are created in a mobile app or after a sales call, website key events undercount them.
- Companies without access. No opt-in, nothing to read. For outside traffic estimates, a third-party estimates tool is the better fit.
- Long look-backs. Comparing this spring's cohort with last spring's at the same age is out of reach, because an agent query stops at one year back. Search history has its own ceiling too: Google's Queries report help page puts Search Console retention at sixteen months.
- Alerts. Nothing pings you the moment a company drops. The closest thing is a recurring email: the agent can set one up with
create_scheduled_report, covering up to 50 websites per report, and it always lands in the platform account's own inbox rather than any founder's. A batch of fourteen fits in one; it still needs someone to read it before Thursday. - One-call portfolio answers. The agent reads each website separately and combines the results, paging through websites 100 at a time.
Later-stage holdings need a different playbook; see our page for private equity operating partners, or the portfolios overview for quarter-on-quarter views.
Founder objections, and honest answers
What exactly will the platform team see about my company?
Daily sessions and signups (your GA4 key events), traffic by channel, top landing pages, and Google search clicks, impressions, average position and top queries. Totals, not individual visitors. No revenue, product, CRM or billing data, because DuckDive does not read those systems.
Can anyone at the fund change our Google Analytics or Search Console?
Not with the access this page recommends. A read-only role reads reports and cannot change your setup. DuckDive's own actions, such as adding a website, happen inside the platform team's DuckDive account, and nothing the connector does can erase a founder's data.
How do I withdraw, and what happens to data already pulled?
Remove the platform login in Google Analytics under Admin, Access Management, and in Search Console under Settings, Users and permissions. New data stops arriving from then on. History imported before that stays in the platform team's DuckDive account until they remove the website, so ask them to confirm in writing when they have.
Will other founders in the batch see our numbers?
Not through DuckDive unless the platform team gives them access. A connected agent sees only the websites the connecting account can see. Whether the team shares a batch table is its own policy, and a fair opt-in note says so.
Does our analytics data end up with an AI company?
DuckDive does not receive the conversation between the platform team and its assistant. When the agent reads your numbers, the tool results go to that assistant, and the AI provider's own policies govern them. It is fair to ask which assistant and plan the team uses.
We have no GA4. Do we have to install it to take part?
No. Search Console alone covers the search signals, including searches for your own name. If you add GA4 later, Google says data collection can take up to 30 minutes to begin once the tag is live, and reports start from that day.
Does this cost the startup anything?
No. The startup grants access with the Google accounts it already has. Any DuckDive plan sits with the account that connects the sites, which is the platform team's, and the first site is free.
Sources
- Google Analytics Help: Add, edit, and delete Analytics users and user groups. Opened and checked on November 16, 2026.
- Search Console Help: Managing owners, users, and permissions. Opened and checked on November 16, 2026.
- Google Analytics Help: About key events. Opened and checked on November 16, 2026.
- Google Analytics Help: [GA4] Recommended events. Opened and checked on November 16, 2026.
- Google Search Central Blog: Introducing the branded queries filter in Search Console. Opened and checked on November 16, 2026.
- Google Search Central Blog: A deep dive into Search Console performance data filtering and limits. Opened and checked on November 16, 2026.
- Search Console Help: Add a website or platform property to Search Console. Opened and checked on November 16, 2026.
- Google Analytics Help: Set up Analytics for a website and/or app. Opened and checked on November 16, 2026.
- Google Analytics Help: [GA4] Queries report. Opened and checked on November 16, 2026.
Last reviewed . Numbers in EXAMPLE DATA blocks are fictional, and so are the companies.
