·8 min read·
Notes
Workflow
Guide

Search Console annotations: log SEO changes, then see what worked

Google Search Console annotations mark a date on one chart. Log SEO changes as notes your AI assistant reads, then measure before and after. With prompts.

Search Console annotations are short dated notes you pin to the Performance report chart in Google Search Console: right-click the chart, pick a date, type up to 120 characters, click Add. They are good for seeing when something happened. They do not tell you whether the change worked, they vanish in comparison mode, and no API exposes them, so an AI assistant reading your Search Console data never sees them. This post covers what Google's feature does as of November 2026, and a log-then-measure workflow that uses notes in the Search Console MCP server so Claude or ChatGPT can read your change log and compare the weeks before and after each change.

What Google Search Console annotations do (and don't)

Google launched custom annotations in the Performance report on 17 November 2025, after a test that started in May. Per the Search Console help page and Google's launch post:

  • Two kinds. System annotations are added by Google for data and reporting issues. Custom annotations are yours.
  • Adding one. Open the Performance report, right-click the chart, choose the date, type the note, click Add. Since December 2025 you can annotate today's date, but not a future one.
  • Limits. 120 characters per note, 200 per property. Annotations older than 500 days are deleted automatically. There is no edit: delete and re-add.
  • Who sees them. Everyone on the property. Owners and full users can add and delete any annotation; restricted users can only view.
  • Where they show. On the Performance chart at their date, whatever filters are on. Not in comparison mode, and not in the 24-hour view.

Two gaps matter for change tracking. First, the comparison view is exactly where you judge a change, and the annotations are hidden there. Second, the Search Console API covers search analytics, sitemaps, sites and URL inspection, and nothing else. No annotations endpoint exists, so no tool built on the API (ours included) can read or write them. If you use both, the same fact lives in two places, and you copy it by hand.

Use them anyway. They are free and visible to anyone who opens the report. Just don't expect them to answer "did it work".

Notes in the MCP: a change log your assistant reads

The Search Console MCP server has three note tools: add_note, list_notes and delete_note. A note is plain text, stored on mcpsearchconsole.com (never written to Google), either scoped to one property or account-wide. What changes compared with a chart marker:

  • The assistant reads them before it analyses. The list_notes tool tells the assistant to call it before analysing a traffic change, and list_sites mentions how many saved notes exist. So when you ask "why did clicks drop in October", a well-behaved assistant checks your notes first and says "you moved /blog to /articles on 12 October" instead of inventing a cause. Most do this unprompted; adding "read my notes first" to the prompt makes it certain.
  • Room for the useful detail. Up to 2,000 characters per note, 200 per account. Enough for "rewrote titles on 14 product pages, list below, old titles kept in the CMS history".
  • Not only events. Notes also hold standing context: which queries count as brand, which section is seasonal, a baseline snapshot of last month's totals. The assistant uses that to filter brand queries out or to stop flagging December as a drop.
  • Shared on the Team plan. Everyone on the team sees every note, and each one names its author, so "who changed the title tags" survives. Members delete their own notes; the team owner can prune any.
  • Free. Note calls are not data pulls and don't count toward the free plan's caps.

One detail to get right: list_notes shows the date a note was saved. If you log a change three days late, put the real date in the text: "2026-11-03: shipped new category template". The measuring step below relies on that date.

What to log

A common way to use notes is one fact per note, written the day it happens. The changes worth logging are the ones you will want to measure or explain later:

  • Migrations and redesigns. Domain moves, URL structure changes, a new template, a CMS switch. Pair these with the site migration checks.
  • Content changes. Rewritten titles and meta descriptions, merged or pruned articles, a refreshed guide. Name the URLs or the URL pattern so the assistant can filter to them.
  • Technical changes. Robots.txt edits, noindex added or removed, internal linking changes, a hosting move.
  • Standing context. Brand terms, seasonal sections, known tracking gaps.
  • Baselines. Before a big change, ask the assistant to save the current numbers for the affected pages as a note. Then nothing depends on remembering what "normal" was.
Log a change with a baselineFor sc-domain:example.com: I rewrote the title tags on every page under /guides/ today (2026-11-16) to lead with the main keyword instead of the brand. First call search_analytics for the last 28 complete days, filtered to pages containing /guides/, by page, and give me clicks, impressions, CTR and average position per page. Then save one note with add_note scoped to this property: "2026-11-16: rewrote titles under /guides/ (keyword first, brand last). Baseline last 28 days:" followed by the totals and the five biggest pages with their CTR and position.
Save standing contextSave these as separate notes for sc-domain:example.com with add_note. 1: "Brand queries: example, example app, examplehq. Exclude them when judging SEO changes." 2: "/events/ is seasonal: traffic peaks in March and September, low in December. Do not report its December drop as a problem."

Measure the change: before and after with compare mode

search_analytics has a compare option. Give it a window with explicit start and end dates, and it compares that window with the immediately preceding window of the same length, row by row: clicks, impressions, CTR and position, each with the previous value next to it. For a change shipped on day D, set the window to start on D and run for N days; the comparison period is then the N days before D. That is a clean before and after.

Three rules keep the result honest:

  • Wait. Google has to recrawl the changed pages, and Search Console data trails by about two days. Two weeks after a title change is the earliest useful read; four weeks is better. After a migration, give it longer.
  • Filter to what changed. Compare the affected pages, not the whole property, or a sitewide trend drowns the effect. Exclude brand queries if a note lists them.
  • Use equal windows that avoid known noise. A window that crosses a holiday or a Google core update is not a clean test. Ask the assistant to say so when it does.
Did the change work?For sc-domain:example.com: call list_notes and find the note about rewriting titles under /guides/. Use the date in that note as the change date. Call search_analytics with compare on, startDate set to the change date and endDate 27 days later (or the latest complete day if sooner), dimension page, filtered to pages containing /guides/, excluding the brand queries listed in my notes. Tell me per page whether CTR and clicks moved, and whether position changed (a CTR gain at the same position is the title working; a gain that comes with a better position may be something else). Compare with the baseline numbers in the note. Finish with a one-line verdict and save it as a note starting "Result" with today's date.
Explain a traffic changeClicks on sc-domain:example.com fell over the last two weeks. Before you look at any data, read my notes with list_notes and list every logged change from the last 60 days. Then run search_analytics with compare on for the last 14 complete days by page, and match the biggest losers to those changes. Separate drops you can tie to a logged change from drops nothing in the notes explains, and for those, run inspect_url on the top three affected pages.

The second prompt is where notes earn their keep. Without them, the assistant sees a drop on /articles/ and a matching loss on /blog/ and has to guess. With a migration note, it reports a redirect working as planned and moves on to the drops that need attention. For slow declines that no change explains, the content decay use case is the better playbook; for title work, start from low-CTR pages.

Keep both logs, briefly

A setup that works: the native annotation as a short pointer on the chart ("Titles rewritten /guides/"), the MCP note with the detail and baseline. The chart marker helps anyone who opens Search Console; the note is what your assistant reads. Since the MCP cannot write annotations to Google, you add the chart marker yourself. It takes ten seconds.

If you run a scheduled weekly check, the run reads the same notes, so a logged migration stops showing up as a regression every Monday. The weekly SEO report can include a "changes this week and what they did" section built from notes.

Limits

  • No access to Google's annotations. No public API exists for them, so the assistant cannot read your chart annotations or create them. Copy anything important into a note.
  • Notes are text, not markers. They don't draw on the Search Console chart. The search_chart tool draws a daily trend, but notes are not plotted on it.
  • The assistant decides when to read. The tool descriptions ask it to call list_notes before analysing changes, and most assistants do. It is not guaranteed; say "read my notes first" when it matters.
  • Before and after is not proof. Compare mode shows what moved, not why. Seasonality, a core update or a competitor can move the same pages in the same weeks. Treat a result as evidence, and check positions alongside clicks.
  • Data limits are Google's. About two days of lag, anonymised queries missing from query-level rows, and the 16 months of history the API keeps.
  • Caps. Notes are 2,000 characters each, 200 per account; prune old ones with delete_note. Notes are free, but each search_analytics call is a data pull: the free plan covers 1 property with 15 pulls a day and 40 a month, Solo (4.99 € a month) covers 5 properties without caps, and Team (29.99 € a month) covers unlimited properties for everyone on your company domain, with shared notes.
  • Write approval in ChatGPT. add_note is marked as a write, so ChatGPT may ask you to approve it. Reads are marked read-only. Sign-in only grants read-only access to Search Console itself.

Setup takes a few minutes: connect Search Console to Claude or to ChatGPT, then log the next change you ship.

Ready to put your search data in your AI?

Set up the Search Console MCP, free

Free plan included. See pricing