usdigitalmarketingindex.com SEO
What should an SEO change log record to support later investigations?
Updated November 15, 2026
Enough for someone, months later and with no memory of the event, to answer five questions about any change: what exactly was changed, on which URLs, when it went live, who did it and why, and how it was confirmed. An investigation into a traffic movement works backward from a date. It asks what happened on the site around then. If the only records are a developer’s commit history, a marketing calendar, and people’s recollections, the question cannot be answered with confidence, and causes get assigned by guesswork. A change log puts every change that could affect search into one dated list, with the affected pages named and the evidence attached. Its value is realized long after each entry is written, so the entries have to stand on their own.
Who changed what and when
The core of an entry.
| Field | Content | Why |
|---|---|---|
| ID | A unique reference | To cite the change elsewhere |
| Date and time live | When the change reached production, with time zone | Investigations align on the production time, not the day work was finished |
| Author | Who made the change | Whom to ask |
| Approver | Who signed it off | Accountability |
| Type | Content, template, technical, redirect, robots, sitemap, structured data, internal linking, tracking, infrastructure | Filtering |
| Description | What was changed, in specific terms | The substance |
| Before and after | The old value and the new value | Makes the change reversible and comparable |
| Reason | Why it was done | Context for judging the outcome |
| Reference | Ticket, pull request, or document | Trace to the detail |
Be specific in the description. “Updated titles” records nothing. “Changed the title pattern on product pages from ‘Name | Brand’ to ‘Name: key specification | Brand'” can be evaluated later.
Record the time of deployment, not of the decision. A change approved on Monday and released on Thursday affected search from Thursday.
What belongs in the log:
- changes to page templates, titles, headings, and body content at scale;
- URL changes and redirects;
- robots.txt, meta robots, and canonical changes;
- sitemap changes;
- structured data changes;
- navigation and internal linking changes;
- site speed, hosting, CDN, and platform changes;
- analytics and consent changes, which alter the measurements;
- content published, removed, or merged in bulk;
- Search Console actions: removals, change of address, disavow uploads;
- outages and incidents.
Also log external events that explain data: confirmed search updates, marketing campaigns, press coverage, and seasonal peaks. They are not changes to the site, and they are what an investigator needs beside the changes.
A log records who made a change. Who is responsible for keeping each page correct between changes is a separate assignment, covered in the guide to assigning individual owners responsibility for SEO content maintenance.
Affected URLs or templates
Scope is what turns an entry from a note into evidence.
| Field | Content |
|---|---|
| Scope type | Single URL, list of URLs, template, section, whole site |
| URLs | The list, or a link to it, for anything short of a template |
| Template or pattern | The template name and URL pattern |
| Count | Number of pages affected |
| Exclusions | Pages of the type that were not changed |
| Environment | Production, with the hostname |
| Markets or languages | Which versions |
An investigation asks whether the pages that moved are the pages that were changed. Without scope, a change can be neither linked to a movement nor cleared of it.
For changes applied to a subset, such as a test on 50 pages, attach the list of pages changed and the list left unchanged.
Expected impact and rollback
Record the intention before the result is known.
| Field | Content |
|---|---|
| Hypothesis | What the change was expected to do, and to which metric |
| Metric and source | How it will be judged, and in which tool |
| Review date | When the outcome will be checked |
| Risk | What could go wrong |
| Rollback method | How to undo it |
| Rollback trigger | The condition under which it will be undone |
| Dependencies | Other changes released with it |
Writing the expectation first guards against explaining any outcome as intended afterward. It also records concurrent changes, which is what an investigator most needs to know: if three things shipped in one release, an effect cannot be assigned to one of them without more evidence.
Observed verification
The last part closes the entry.
| Field | Content |
|---|---|
| Deployment evidence | How it was confirmed live: a fetched page, a response header, a crawl, with date |
| Verified by | Who checked |
| Verification URLs | The exact URLs checked |
| Observed state | What was seen |
| Outcome at review date | The metric's movement, with the comparison used |
| Conclusion | Kept, reverted, or modified; confidence stated |
| Follow-up | Next action, if any |
A change is not logged as live because it was merged. It is logged as live when someone fetched a production URL and saw it.
A change-log schema with one fully hypothetical row:
| Field | Entry |
|---|---|
| ID | SEO-0142 |
| Date and time live | 2026-03-12 10:40 UTC |
| Author | Developer A |
| Approver | SEO lead B |
| Type | Template: structured data |
| Description | Added Product structured data with offers to the product page template |
| Before | No Product markup |
| After | Product with name, image, description, sku, offers (price, priceCurrency, availability) |
| Reason | Make product pages eligible for product rich results |
| Reference | Ticket WEB-881; pull request 2207 |
| Scope type | Template |
| Template or pattern | product-detail; <!–INLINECODE0–> |
| Count | 1,840 pages |
| Exclusions | Discontinued products (archive template) |
| Environment | Production, www.example.com |
| Hypothesis | Valid Product items reported in Search Console; possible rise in CTR on product queries |
| Metric and source | Product snippet report valid item count; CTR for product pages in the performance report |
| Review date | 2026-04-23 |
| Risk | Price in markup diverging from visible price on variant pages |
| Rollback method | Feature flag off |
| Rollback trigger | Errors on more than a small share of items, or any price mismatch found |
| Dependencies | Released alone |
| Deployment evidence | Rich Results Test on three product URLs, 2026-03-12 11:05 UTC, results saved |
| Verified by | SEO lead B |
| Verification URLs | /products/thermostat-t200/, /products/radiator-valve-15mm/, /products/boiler-filter-22mm/ |
| Observed state | Product item detected, no errors, price matches page on all three |
| Outcome at review date | To be completed |
| Conclusion | To be completed |
| Follow-up | Check variant pages for price consistency in week 1 |
Every name and number in that row is invented.
Practices that keep the log usable:
| Practice | Reason |
|---|---|
| One log, shared by SEO, development, content, and analytics | Changes from every team in one place |
| Entry made at release, as part of the release checklist | Entries written later are incomplete |
| Read access for everyone; edit history kept | Trust in the record |
| Dates in one time zone | Alignment with data |
| Linked to annotations in analytics and reporting tools | The change is visible on the charts |
| Reviewed monthly for missing entries, against the deployment history | Gaps found while people remember |
| Outcomes filled in at the review date | The log becomes a record of what worked |
| Never edited to remove a mistake; corrections added as new entries | The history stays honest |
A log does not prove causation. It shows what coincided with what. Its use in an investigation is to supply candidates and to rule out dates on which nothing changed.
The log records what people did on purpose. Changes nobody intended are found by another method, comparing the site before and after a release, described in the guide to how two site crawl snapshots can reveal unintended changes after a code release.
The test for an entry: a stranger reading it a year from now can say what changed, where, when, why, and how it was verified, without asking anyone.
Add your comment to this answer