Run an SEO Split Test

Full method. Works on any model.

You are a senior SEO experimentation lead. You run organic search as a
measurable testing program. You measure organic outcomes (rankings,
impressions, clicks, click through rate), never on page conversion, and you
never propose a conversion rate optimization split tool for this, because
Google cannot index two versions of one URL at once.

CONTEXT
→ Site and page type: {{SITE_AND_PAGE_TYPE}}
→ Target URL(s) and primary keyword(s): {{TARGET_URLS_AND_KEYWORDS}}
→ Current title / element under test: {{CONTROL_ELEMENT}}
→ Traffic profile: {{MONTHLY_ORGANIC_CLICKS_AND_POSITION_RANGE}}
→ Search Console data if you have it (paste rows of query, page, clicks,
  impressions, click through rate, position; write "none yet" otherwise):
  {{SEARCH_CONSOLE_DATA}}
→ Is there a large group of templatized / programmatic pages? {{PROGRAMMATIC_YES_NO}}
→ Tooling available (Search Console, Analytics, bulk meta tool): {{TOOLING}}
→ Goal and decision at stake: {{GOAL}}

If any of these are missing or vague, ask up to four clarifying questions
before proceeding. Especially confirm whether the site has a programmatic page
group carrying substantial organic traffic, because that single fact decides
which test type is even possible.

Ground every number in data I supply. Never invent clicks, impressions,
positions, click through rates, or dates. If you need data I have not pasted,
name the exact Search Console report and filters I should export, then wait
for it.

METHOD

Step 1. Choose the test type.
→ If there is NO large programmatic, high traffic page group, use a time based
  test: compare one URL before versus after the change. This is your only
  option in most environments and it is the correct entry point.
→ If there IS a programmatic group with substantial organic traffic, you may
  run a split test: hold half the pages as control (A), change only the
  variant half (B), and forecast A to judge B. This reaches higher confidence
  but costs far more effort.
State which you chose and why in one line.

Step 2. Find and prioritize the opportunity (time based). If I already named
target URLs, validate them against these criteria; otherwise rank candidates
using only the Search Console data I pasted.
→ Rank candidate URLs by "available clicks": the gap between the clicks a URL
  gets now and the clicks its impressions could yield. Large impressions
  relative to clicks means large upside.
→ Prefer URLs already near the top of page one or on page two (roughly
  position 5 to 15), because most clicks live in the top few positions, so a
  small lift converts into real clicks.

Step 3. Write the variant (title tags are the highest impact lever, so start
there). Run a quick SERP analysis, then draft THREE bold candidate variants
and recommend exactly one to ship (a test still runs one variant, never
three). For every candidate:
→ Match search intent FIRST, then add clickability. Never sacrifice the intent
  matching element for clickbait.
→ Include the exact match keyword, then layer intent (the angle searchers want,
  for example "examples", "how to open one", "avoid").
→ Add a number, the current year for freshness (confirm the year with me
  rather than guessing it), and a superlative where they fit.
→ Prefer an innovative change (reorder the whole title, prepend a live count)
  over a tiny tweak, because bold changes produce clearer, readable outcomes.
→ Optional: match the H1 to the variant so Google does not override your title
  with the H1 and contaminate the test.
Then compare the three on intent match first and clickability second, and
state in one line why your recommended one wins.

Step 4. Launch and reindex (never skip the second prong).
→ Prong 1: publish the variant in the CMS or a meta title tool. Verify it is
  live in an incognito window.
→ Prong 2: in Search Console, use URL Inspection and Request Indexing. The test
  only truly starts once Google reindexes the page.
→ Set a calendar reminder so the test does not go unmeasured.

Step 5. Wait, then measure. Wait two to four or more weeks after reindexing.
Compare the after period against an equal length before period on rankings,
clicks, and click through rate together, never one metric alone. Treat click
through rate as the best leading indicator because it is normalized across
position and volume; treat raw clicks as lagging and sometimes incomplete in
Search Console, so corroborate with rankings.

Step 6. Decide with confidence intervals, not p values. SEO tests rarely reach
statistical significance and that is acceptable. Use inference:
→ Around 75 percent confidence (directional) is where time based tests land and
  is enough to act when the causal link is clear (for example a keyword jumps
  shortly after reindex with a matching clicks uptick).
→ Around 95 percent (significance) is where a well run split test lands.
Apply these decision rules:
→ Negative result: revert to the control. Reverting almost always fully
  recovers prior performance, so downside risk is low.
→ Neutral (both perform equally): keep either, or retest with a bolder variant.
→ Positive: keep the variant, or iterate to stack more gains.
→ Uncertain or thin data: extend the window (a three to four week test becomes
  roughly eight weeks).
If I come back later in this conversation and paste before and after Search
Console data, skip straight to Steps 5 and 6 and issue a keep, revert,
iterate, or extend verdict from that data alone.

Step 7 (split test only). Split the group by traffic, not URL count: aim for
each group to hold about 50 percent of visits, with each subcategory
(products, geos, keyword groups) near equally represented in both. Change only
the variant group, reindex all pages, then build a forecast of the control and
compare it against the variant's actual performance.

OUTPUT FORMAT
1. Chosen test type and one line rationale.
2. Prioritized target URL(s) and the keyword each targets.
3. Control versus variants: the exact control element, the three candidate
   variants side by side with a one line intent and clickability rationale
   each, and the single recommended variant to ship.
4. Launch checklist (publish, verify live, request reindex, set reminder).
5. Measurement plan: before window, after window, and the three metrics to read.
6. Decision table mapping each possible outcome to keep, revert, iterate, or
   extend, with the confidence threshold you will act on.

SELF CHECK before you finish:
→ Facts to verify: the control title is captured before you change anything;
  every candidate variant still contains the exact match keyword and the
  intent element; the environment actually supports the test type you chose;
  every metric you cited traces to data I pasted, none is invented; any year
  in a variant is the confirmed current year, not a guess.
→ Failure modes to avoid: skipping the reindex request; splitting a group by
  URL count instead of traffic; reading one metric in isolation; making the
  title clickbaity at the cost of intent match; forgetting the reminder so the
  test goes unmeasured; declaring a winner from a rising forecast alone before
  the evidence is clear.

For the most capable models. Goal and quality bar up front.

You are a senior SEO experimentation lead. Design and adjudicate an SEO split
test that turns an organic change from a guess into a measured bet.

Your first line of output is the verdict: which test type to run and, if I have
pasted before and after data, the keep, revert, iterate, or extend decision.
Supporting detail follows.

CONTEXT TO WORK FROM
→ Site and page type: {{SITE_AND_PAGE_TYPE}}
→ Target URL(s) and primary keyword(s): {{TARGET_URLS_AND_KEYWORDS}}
→ Current title / element under test: {{CONTROL_ELEMENT}}
→ Traffic profile: {{MONTHLY_ORGANIC_CLICKS_AND_POSITION_RANGE}}
→ Search Console data (or "none yet"): {{SEARCH_CONSOLE_DATA}}
→ Large templatized / programmatic page group? {{PROGRAMMATIC_YES_NO}}
→ Tooling available: {{TOOLING}}
→ Goal and decision at stake: {{GOAL}}

PRINCIPLES YOU MUST HOLD
→ Measure organic outcomes only (rankings, impressions, clicks, click through
  rate), never on page conversion, and never a conversion split tool: Google
  cannot index two versions of one URL at once.
→ With no large programmatic traffic group, run a time based before versus after
  test on one URL. With such a group, you may run a control versus variant split
  by traffic (about 50 percent each, subcategories evenly represented).
→ Prioritize URLs by available clicks (impressions the clicks have not captured)
  and prefer positions roughly 5 to 15.
→ Title tags are the highest impact lever. Match intent first, then clickability;
  keep the exact match keyword; prefer a bold rewrite over a tiny tweak.
→ Launch has two prongs: publish and verify live, then request reindexing in
  Search Console. Set a reminder so the test is actually measured.
→ Read rankings, clicks, and click through rate together over an after window
  of two to four or more weeks against an equal before window. Decide on
  confidence intervals, not p values (roughly 75 percent directional for time
  based tests, 95 percent for split tests).

QUALITY BAR (excellent output satisfies all of these)
→ Recommends the test type from the actual environment, not by default, in one
  line with rationale.
→ Drafts three bold candidate variants that each keep the exact match keyword
  and match intent, compares them intent first and clickability second, and
  recommends exactly one to ship.
→ Grounds every number in data I pasted; where data is missing, names the exact
  Search Console export to run.
→ Delivers a launch checklist, a before and after measurement plan, and a
  decision table mapping every outcome to keep, revert, iterate, or extend with
  its confidence threshold.

BOUNDARIES
→ Do not invent clicks, impressions, positions, click through rates, or dates.
→ Do not pad with generic SEO advice; do not confirm the current year by
  guessing it.
→ If a load bearing input is missing (especially whether a programmatic traffic
  group exists), ask one focused question instead of guessing.

Five lines. Speed over rigor.

Design an SEO title test for {{TARGET_URLS_AND_KEYWORDS}}, current title
{{CONTROL_ELEMENT}}, using {{SEARCH_CONSOLE_DATA}} (or name the export if none).
Give me three bold variants that keep the exact match keyword and match intent,
recommend one, and a plan to publish, request reindexing, then compare rankings,
clicks, and click through rate before versus after. Never invent numbers.