For content site portfolios

Decide which content sites to grow, hold, prune or sell with your agent

You run a stack of content and affiliate sites. Every quarter you guess where the writers go and which site to list. Put three Search Console numbers behind that call, pulled by your agent for every site.

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 18 min read

The short answer

Score every site on three Search Console numbers: how its last 6 months of clicks compare with the 6 before, the share of query clicks from its top 10 queries, and the share of all clicks landing on its single best page. Those three place each site in grow, hold, prune or sell. DuckDive stores each site's history so your AI agent can score the whole portfolio.

  • Check one-page dependency first. A growing site that hangs on one URL is not a growth bet.
  • Fund new articles where growth is broad; give falling sites editor time.
  • Earnings are not a data source. Paste your own and let the agent divide them by GA4 sessions.

Grow, hold, prune or sell: the four-box matrix

Most portfolio owners already sort their sites like this in their heads, from memory: the site that had a good month, the one a favourite writer covers, the one you keep meaning to fix. Put each site in a box using Search Console numbers instead, and the quarter's calls get easier to defend.

Grow

Last 6 months' clicks up 10% or more on the 6 before, no page above 25% of clicks, top-10 query share under 50%.

Gets most new articles. Growth is spread out, so each piece lands beside pages that already earn.

Hold

Everything that is not growing broadly, not falling hard and not hanging on one page.

Refreshes only. Keep earning pages current and fix the ones slipping.

Prune

Last 6 months' clicks down 15% or more on the 6 before, with no page at 40% or above.

No new articles. Editor time goes to merging, rewriting or removing pages that no longer help a reader.

Sell

One page earns 40% or more of the site's clicks, whatever the trend.

Spread the risk with supporting pieces, or list it while that page still ranks.

Test the boxes in this order: Sell, then Prune, then Grow, with Hold catching the rest. A growing site with one dominant page is still a one-page site, which is why Sell goes first. These cut lines are a starting point chosen for this page, not an industry standard. Tighten them if you have more sites than writers.

By hand, the inputs are tedious: two click totals, a query list and a page list, times every domain you own, with a property switch in Search Console for each. Linking a site makes DuckDive import its Search Console and GA4 history and hold on to it, and your AI agent reads that store through a remote MCP server. The agent reads one site at a time from history already held, then does the sums in the chat. Stored history versus live API servers explains why that matters at portfolio scale.

Three numbers that place a site in a box

Each number answers one question a buyer or partner would ask. All three come from Search Console, the only source the sort needs; what DuckDive reads from Search Console lists the fields.

Half-year click trend
Search Console clicks for the last 6 months, divided by (clicks for the last 12 months minus clicks for the last 6 months), minus one. Both inputs are plain window totals, so the agent needs no monthly series to work it out. Google counts a click each time someone clicks a link from its results to your site.
Top-10 query share
Clicks from the site's 10 biggest queries, divided by clicks from its top 100 queries, over the last 90 days.
Single-page share
Clicks to the site's best page, divided by the site's total clicks, over the last 90 days.
Average position
The average position of the topmost result from your site, as Google defines it. A tie-breaker, not a placing number.

Top-10 query share is computed from the top 100 queries only. The query tool returns up to 100 rows, so the denominator is what those 100 queries earned, not every click on the site. That is deliberate. Search Console leaves rare "anonymized queries" out of its query table but counts them in the chart total, per Google's help on performance report dimensions, so dividing a query sum by the site total mixes two bases. The cost: the share reads higher than "share of all clicks", most on long-tail sites. Compare your sites with each other, not with outside figures.

Single-page share has a smaller wrinkle: page rows are aggregated by page and the chart total by property, per Google's performance report overview. It rarely changes the box.

Seasonal niches

A gardening or ski site compares its busy half-year with its quiet one, and the trend measures the calendar. The agent reaches back 12 months, so it cannot set a month beside the same month last year. Search Console keeps 16 months, per Google's Analytics help on the Queries report, so check like months there first.

Ask your agent
For every site, give me total Search Console clicks for the last 12 months and for the last 6 months, the trend between the two halves, and the share of clicks coming from its top 10 queries.

Tools it calls list_websites get_search_console_summary get_search_console_keywords

Worked example: an eighteen-site portfolio sorted into boxes

Here is the whole sort on a fictional portfolio. The cut lines go in the prompt, so the agent applies your rules instead of inventing its own. Keep it for every quarter.

Ask your agent
Sort every site I own into grow, hold, prune or sell. Trend: last-6-month clicks divided by (last-12-month clicks minus last-6-month clicks), minus one. Concentration: last 90 days. Sell if one page has 40%+ of clicks. Prune if trend is -15% or worse. Grow if trend is +10% or better, no page above 25%, and top 10 queries under 50% of top-100 query clicks. Hold is the rest. Show each site's three numbers.

Tools it calls list_websites get_search_console_summary get_search_console_keywords get_search_console_pages

EXAMPLE DATAFictional business: an eighteen-site publisher, including Fernhollow Gardening

Run that on all my sites and group them by box.

  1. list_websitesfinds the 18 sites and checks each has Search Console linked
  2. get_search_console_summaryeach site, click totals for the last 12 months, the last 6 months and the last 90 days
  3. get_search_console_keywordseach site, top 100 queries by clicks, last 90 days
  4. get_search_console_pageseach site, its top pages by clicks, last 90 days

Five sites land in Grow, five in Hold, four in Prune and four in Sell. Nettlecombe Beekeeping grew 14%, but its top 10 queries carry 57% of query clicks, so it stays in Hold. Cindertrack Running has not fallen 15%, yet one page earns 52% of its clicks, so it goes to Sell.

SiteTrendTop 10Top page
Grow
Kettlerock Climbing+26%38%14%
Brookmere Dog Training+21%27%11%
Fernhollow Gardening+18%31%9%
Driftmark Kayaking+15%41%22%
Moonwake Astronomy+12%44%19%
Hold
Nettlecombe Beekeeping+14%57%28%
Otterbrook Fly Fishing+9%52%21%
Thimblewick Sewing+4%35%12%
Heronfield Birding-2%29%8%
Hollowreed Guitar-6%48%17%
Prune
Stonefly Tabletop-17%33%15%
Lanternfly Camping-19%39%13%
Fogwillow Aquariums-24%46%18%
Pebbledash DIY-31%22%6%
Sell
Pinecastle Kitchen+7%64%47%
Tidepool Travel Notes+11%58%41%
Gearloft Reviews-4%71%43%
Cindertrack Running-12%76%52%

Pebbledash DIY falls fastest, but no page holds more than 6% of its clicks. That is slow decay across the site, not one lost ranking, so it is a pruning job.

Search Console

Two of the four Sell sites are growing. A monthly traffic report would have praised both.

Sort your own sites before the next writer brief

Link each site's Search Console, paste the sorting prompt, and see which of your domains are quietly living on one page.

Continue with Google

Free for 2 sites · No credit card

Single-page dependency, and why buyers discount it

A site whose traffic rests on one page is one ranking change from being a different business. Google's page on core updates and your website says ranking changes "can cause both gains and drops in organic Search traffic", and that after a site improves "it could take several months" for its systems to confirm it. A deep bench of earning pages rides that out. One URL holding half the clicks may not.

Buyers look for it. FE International's guide to advertising due diligence says "it can be a red flag if one page is accounting for a disproportionate amount of traffic", and tells buyers to ask the seller why. Flippa's article on one site versus several says the same from the owner's side: "One website with 100 high ranking pages is still probably safer than 3 websites with a handful of popular articles each." If you list a Sell-box site, expect a buyer to ask about its top page early.

FE also suggests asking whether the page is the homepage. Our own reading, not FE's: a top page held up by searches for the site's own name is sturdier than a recipe ranking for a generic phrase, because a rival cannot easily outrank you for your own name.

Ask your agent
Which of my sites gets more than 40% of its clicks from a single page?

Tools it calls list_websites get_search_console_pages get_search_console_summary

Three ways to lower the share before a sale

  • Build around the winner. Write what readers of the big page search for next, and link it from there. Every click those pieces earn lowers the share.
  • Protect the winner first. Refresh it before anything else. A dip during a sale shows up in the latest 90 days, the window buyers read hardest.
  • Time the listing to the direction. A share falling from one 90-day window to the next tells a better story than one that is high and flat.

On the other side of the deal, verifying a seller's traffic before you buy covers access and authenticity checks, and uses this concentration test.

Where next quarter's writing budget should go

The matrix says which sites deserve articles, not which topics. For topics, brief writers on queries already rising on the Grow sites.

Ask your agent
On Kettlerock, Brookmere, Fernhollow, Driftmark and Moonwake, which of the top 100 queries gained the most clicks in the last 90 days compared with the 90 days before? Group them by site so I can turn them into writer briefs.

Tools it calls get_search_console_keywords

There is no ready-made "previous quarter" range, so the agent builds one: each query's clicks over the last 6 months, less its clicks over the last 90 days, leaves roughly the 90 days before (about 92 days, close enough for a brief). A rising query that appears in the 90-day list but not in the 6-month one has nothing to subtract from, so treat it as new rather than as a percentage gain.

Do not spread the budget thin so every site gets something. Google's guidance on helpful, reliable, people-first content lists as a warning sign "producing lots of content on many different topics in hopes that some of it might perform well in search results". Depth on fewer sites serves readers better.

Prune sites get editor time, not writer time. The same page asks whether you add or remove content mainly to make a site seem fresh, and answers: "(No, it won't)". Remove a page because it no longer helps a reader.

Ask your agent
For Pebbledash, Lanternfly, Fogwillow and Stonefly, list the pages from the top 100 that earned less than half as many clicks in the last 90 days as in the 90 days before. Those are my merge or rewrite candidates.

Tools it calls get_search_console_pages

EXAMPLE DATAHow a 60-article quarter splits across the eighteen fictional sites
BoxSitesNew articlesRefreshes or rewrites
Grow5506
Hold5015
Prune (merges and rewrites)4012
Sell (Pinecastle 6, Gearloft 4)4104

The split is a choice. The rule is what matters: new articles follow broad growth, Prune sites get rewrites drawn from the candidate list above and never a fresh piece, and Sell sites only get pieces that cut dependency.

Worked example: two sites with equal clicks and unequal risk

On a traffic report these two fictional sites look like twins. Which one earns the next batch of writer time?

Ask your agent
Compare Fernhollow Gardening and Pinecastle Kitchen on 6-month click trend, average position and traffic concentration, then tell me which deserves the next 20 articles.

Tools it calls list_websites get_search_console_summary get_search_console_keywords get_search_console_pages

EXAMPLE DATAFictional business: Fernhollow Gardening and Pinecastle Kitchen

So, Fernhollow or Pinecastle for the next 20?

  1. list_websitesfinds both sites
  2. get_search_console_summaryeach site, clicks and average position for the last 12 months and the last 6 months
  3. get_search_console_keywordseach site, top 100 queries, last 90 days
  4. get_search_console_pageseach site, top pages, last 90 days

Fernhollow Gardening gets the 20. Clicks are tied, but the risk behind them is not.

MeasureFernhollowPinecastle
Clicks, last 6 months48,20048,600
vs the 6 months before+18%+7%
Average position, 12 months vs last 613.5 to 12.98.2 to 8.1
Top-10 query share31%64%
Single-page share9%47%

Fernhollow grew 18% and no page holds more than 9% of its clicks, so new articles join a site that earns across many searches. Its average position is weaker, but the last 6 months beat the 12-month figure, so it is still climbing and has room to go.

Pinecastle ranks higher, yet 47% of its clicks land on one page, /no-knead-sourdough. Twenty unrelated recipes would barely dent that. If you keep it, give it four to six pieces sourdough readers look for next.

One caution: Fernhollow's latest 6 months include the summer peak and the 6 before include winter, so part of the 18% is season. Check the same stretch a year earlier in Search Console before you sign the brief.

Search Console

Clicks were tied and position favoured the loser. Concentration decided it.

Bringing your own earnings into the conversation

DuckDive does not read ad network or affiliate earnings. Of its six sources, a publisher needs two: Google Analytics 4 and Search Console. There is no revenue or RPM feed, and the agent should never guess one.

Paste earnings in instead: export each site's total from your ad and affiliate dashboards and have the agent divide by GA4 sessions for the same window. DuckDive does not receive your conversation, so pasted figures stay with your AI provider under its own policies.

EXAMPLE DATAFictional business: Gearloft Reviews and three sister sites

Earnings for the last 90 days, ads plus affiliate: Gearloft $9,840, Tidepool $4,120, Kettlerock $6,300, Fernhollow $5,560. Divide each by GA4 sessions for the same 90 days and rank them by earnings per 1,000 sessions.

  1. list_websitesmatches the four names you pasted to your sites
  2. get_ga4_overvieweach site, sessions over the last 90 days

Fernhollow Gardening earns the most per visit, and Gearloft Reviews earns about six times what Tidepool Travel Notes earns. Gearloft and Tidepool both sit in Sell, and the earnings push them opposite ways.

SiteSessionsEarnings pastedPer 1,000 sessions
Fernhollow Gardening29,600$5,560$187.84
Gearloft Reviews61,500$9,840$160.00
Kettlerock Climbing72,400$6,300$87.02
Tidepool Travel Notes158,000$4,120$26.08

At Gearloft, the page carrying 43% of clicks also carries your best-paid visits, so cutting that dependency is worth writer time before any listing. Tidepool brings the most sessions and earns the least from each, so a buyer who monetises travel better may value it more than you do. Fernhollow and Kettlerock, both Grow sites, earn well per visit; Fernhollow's lead also backs the earlier call to give it the next 20 articles.

GA4

Match the dates

Ask the agent which dates its 90-day window covered, then export earnings for exactly those days. Affiliate programs differ on whether a commission is dated by the click, the order or the day it locks.

If a Grow site's sessions climb while earnings stall, its new pages may rank without sending readers anywhere that pays: see pages that rank but don't convert. To recompute any division the agent hands you, use checking your agent's numbers.

Connecting a portfolio that was partly bought

Few publishers built every site they own. Some came from a marketplace, some were launched as sister sites with half the tracking, and two domains may still share one Analytics property from a test years ago. Each of those quirks decides which steps a site can take part in, so sort them out before the first quarterly run.

  1. Take Search Console ownership of the sites you bought

    A purchased site's Search Console property often still sits under the seller's Google account, with you added as a user during handover. That arrangement is fragile. Google's help page on managing owners, users and permissions says that if every verified owner is removed, "all remaining users and delegated owners will lose access to the property after a grace period". Verify ownership with your own account first. The same page warns that a removed owner whose verification tokens stay on the site "will be able to re-verify ownership", so delete the seller's tokens (a DNS record, an HTML file or a meta tag, among others) once you hold your own.

    Do this before linking the site, so DuckDive imports through an account you control. If several purchases are pending, the access guide has the request wording to send a seller or a former partner.

  2. Sign in with the account that now owns them

    Use the Google account you just verified on those properties, through the button at the top of this page or through your agent with the prompt below. If you want the agent to add your eighteen or forty domains itself, it has to connect from an admin login with the "Also allow actions" box left on; anything less and it reads but does not set up.

    Paste into your agent
    Set up DuckDive for my website. Follow https://analytics.duckdive.ai/agents.md
  3. Settle a shared Analytics property before you link it

    Old portfolios often have two domains reporting into one GA4 property, say a cooking site and the baking spin-off launched from it. DuckDive attaches each Analytics property, and each Search Console site, to a single website only. Give the shared property to the domain that brings most of its sessions; the other site still sorts on its own Search Console data and joins the earnings step once it has a property of its own. Bulk-adding many domains has its own guide: onboarding dozens of websites with an agent.

  4. Let sister sites without GA4 sort anyway

    A site that never had Analytics installed is not blocked. Its three placing numbers come from Search Console, so it takes its box like any other. It drops out only when you divide pasted earnings by sessions. Install GA4 on it now and keep it off the earnings ranking until a full 90 days has been collected, or its per-visit figure will be computed on a partial window.

  5. Date the first sort

    Once each site's history has imported, run the sorting prompt and keep the answer with the date it was run. Next quarter, ask which sites changed box since then; a site sliding from Grow to Hold matters more than any single number.

Client-specific connection steps: Claude, ChatGPT, Claude Code.

Each box's blind spot

Every box is placed by search numbers alone, and each one has a mistake it is prone to. Know them before you act on a placement.

Sell: the top page might be your brand
The agent receives top queries and top pages as two separate lists, so it cannot say which searches land on the page carrying 47% of clicks. If that page is the homepage and the queries are your site's name, the dependency is loyal readers, not one fragile ranking. Before listing, filter that one page in Search Console's own Performance report and read its queries.
Prune: falling clicks, unknown cause
The box knows clicks fell, not why. Pages dropped from the index, thin articles, a template change and lost links all look the same from here, and DuckDive holds no indexing reports, link data or competitor rankings. Check Search Console's indexing reports, or an SEO crawler, before you delete anything.
Grow: a long tail hides behind the head
Top-10 share is measured inside the top 100 queries. A site earning across thousands of small searches can grow mostly in queries the agent never receives, so its placement rests on the head. On such sites, weigh the trend more than the concentration numbers.
Hold: one year is the longest lens
Up to 2 years of imported history stays in DuckDive, yet no agent range reaches past 12 months today, so a Hold site that swings with the seasons cannot be compared with the same quarter last year through the agent.
Every box: money arrives by paste
No ad network, RPM, EPC or affiliate figures are read from anywhere. The earnings step only knows what you type in.

A publisher with a handful of domains can flip between properties in Search Console and skip the matrix altogether; it starts saving time once a full pass no longer fits in one sitting. For the weekly check between sorts, see running analytics for dozens of websites. When several sites drop in one week, start with diagnosing a traffic drop across many websites. If you hold operating companies, not content sites, the portfolio overview fits better.

Publisher questions about connecting

Does DuckDive pull my ad network or affiliate earnings?

No. DuckDive reads Google Analytics 4, Search Console, Google Ads, Local Services Ads, Yext and Local Falcon, and none of those carry publisher earnings. Paste earnings per site from your own dashboards into the chat, and your agent can divide them by GA4 sessions for the same dates.

My sites are spread across more than one Google account. Can I still connect them all?

Yes, as long as the Google account you sign in with can see each site's Search Console and GA4 property. The simplest route is to add that one account as a user on every property. It is a normal Google permission, granted by whoever owns the property today.

Why does the agent's top-10 query share differ from my own Search Console export?

The agent divides by the top 100 queries it receives, not by every click on the site. Search Console leaves anonymized queries out of its query table while counting them in the chart total, so a share built on total clicks comes out lower. Both are fair. Just compare every site on the same basis.

Can the agent change anything in Search Console or Analytics?

The connector cannot delete data or disconnect your Google account. Reading is the default. Setup actions, such as adding a site or linking its sources, only work for admin accounts that kept Also allow actions ticked when they connected.

Do all my sites need GA4 for the matrix to work?

No. The three placing numbers come from Search Console alone. GA4 matters only when you bring earnings in, because earnings per 1,000 sessions needs a session count. A site without GA4 can still be sorted. It just sits out the earnings step.

Can I test the sort on one site before adding the portfolio?

Yes. One site is free with no credit card, which is enough to check the three numbers against what you already see in Search Console for that site. Add the rest of the portfolio once the agent's reading matches yours.

Sources

  1. Performance report (Search results): Dimensions and data groupings, Search Console Help, Google, checked November 16, 2026.
  2. Performance report (Search results): Overview and basic setup, Search Console Help, Google, checked November 16, 2026.
  3. [GA4] Queries report, Analytics Help, Google, checked November 16, 2026.
  4. Google Search's Core Updates, ranking-change guidance for site owners, developers.google.com, checked November 16, 2026.
  5. Managing owners, users, and permissions, Google's Search Console help on verified owners and tokens, checked November 16, 2026.
  6. Advertising Due Diligence: How to Evaluate Website Advertising, FE International, checked November 16, 2026.
  7. One Site vs Multiple Sites? It Depends, Flippa, checked November 16, 2026.
  8. Creating helpful, reliable, people-first content, Google Search Central, checked November 16, 2026.

Last reviewed . Numbers in EXAMPLE DATA blocks are fictional.

Put next quarter's writer budget on evidence

Connect your sites once, then ask your agent which ones earn new articles and which ones to list.

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