Both start as AI drafts. One gets published untouched. The other gets human editing, fact checks, and real expertise added. This is what actually happened in a real 16 month experiment that tracked both, using live Google data.
Real data from the SE Ranking AI Content Experiment, 2024 to 2026
Raw AI, faded outEdited AI, still ranking
In the experiment, edited AI articles held their rankings for over a year. The raw AI pages collapsed after about 3 months.
01The experiment setup
SE Ranking ran two tests at the same time. Same AI writer. The only real difference was whether a human touched the draft before it went live.1
Test A, raw AI
2,000 AI articles, zero editing
Published across 20 brand new websites
Generated in one click, used as is
No edits, no fact checks, no images
No author, no internal links
Left to rank on merit alone
Test B, AI plus humans
6 AI articles, fully edited
Published on an established blog
Guided with intent, audience, and outline
Revised and fact checked by experts
Parts rewritten from scratch
Clear disclosure that AI was used
02What happened after a year
The gap was not subtle. Here are the real numbers each side ended up with.1
Raw AI, 2,000 pagesCollapsed
0 → 0
Share of pages in Google's top 100 fell from 28% to 3% around month 3, then never recovered through month 16.
~0.7
clicks per page (1,381 total)
0
recovery after the drop
AI plus edits, 6 pagesSurvived
0 of 6 in top 10
Half of the edited articles reached Google's organic top 10 and stayed there. Four were cited as sources in AI Overviews.
0
clicks per page (2,300+ total)
0
impressions in a year
The raw AI traffic curve
2,000 unedited AI pages. A fast rise, then a cliff. This is the shape of scaled AI content with no human layer.1
Pages ranking in Google top 100Raw AIEdited AI
⚠
The trap: raw AI often ranks well for the first 2 to 3 months. In this test, 70 to 75% of all traffic came before the collapse. That early spike fools people into thinking it works.
03Not just one study
A separate large scale analysis found the same pattern at the very top of search.2
Semrush, 20,000 keywordsHuman edge
0x
At Position 1, content classified as human written was 8 times more likely to rank than AI content, across 20,000 keywords and their top 10 results.
The perception gapCareful
0
of SEOs believed AI content ranks as well as or better than human content. The ranking data told a different story near the top.
Note: AI detection tools are not perfect, so treat the exact split as directional. The direction, though, matches the SE Ranking experiment: editing and expertise win over time.
04Try it yourself
Drag the slider to add human editing to a raw AI draft. Watch the article and its quality scores climb.
Editing levelRaw AI draft
Slide right to add real experience, facts, and an author.
Raw AIFully edited
Live preview of your article
Best Budget Espresso Machine (Ultimate Guide)
In today's world, espresso machines are very important. There are many great options to choose from for every budget.
Real experience 0
Original value 10
Trust signals 5
Survival chance 15
05Why the edited one wins
Google judges the result, not the tool. Here is what the human layer adds that raw AI cannot.
What Google looks at
Raw AI
AI plus edits
Real experienceThe extra E in E-E-A-T
Missing
Added by human
Original valueInfo you cannot find elsewhere
Rehashed
Fresh angle
A named authorSomeone who stands behind it
None
Real expert
Facts checkedBefore publishing
Not checked
Verified
Scaled spam riskChance of a penalty
High
Low
06Straight from Google's rulebook
GoogleSearch Central, official guidance
Google does not ban AI. It rewards quality and punishes low effort at scale, no matter who or what wrote it.3
AI is allowed
Google says using AI or automation in the right way does not go against its guidelines.
"Our focus is on the quality of content, rather than how it is produced."
But spam at scale is not
Making many pages with no added value can break the scaled content abuse rule. Google names this directly in its generative AI guidance.
This is exactly what sank the 2,000 raw AI pages.
Show real experience
Google asks who made the page, how, and why. It wants proof a real person with real experience stands behind it.
Raw AI cannot fake first hand experience. A human editor adds it.
No penalty for AI itself
There is no rule that you must disclose AI, and no penalty just for using it. Being open and adding review still builds trust.
Judged on the result, against Search Essentials and spam rules.
E
Experience
E
Expertise
A
Authority
T
Trust, valued most
07The edit pass that saves an AI article
This is the human layer that turned a doomed draft into a survivor. Use it on your own content.
✓
Add real experienceYour own tests, photos, numbers, and mistakes.
×
Publish the raw draftGeneric "in today's world" intros scream low effort.
✓
Check every factAI makes up stats. Verify before it costs you trust.
×
Mass produce pagesBulk pages for rankings is the fast track to a penalty.
✓
Add a real authorName, bio, and credentials Google can see.
✓
Answer the real questionHelp the reader, do not just chase the keyword.
The verdict
SURVIVES
The data is clear. AI writes the draft. A human makes it worth ranking. The article that survives is not the one that hid its AI. It is the one that added what AI cannot: real experience, checked facts, and a person willing to put their name on it.
Sources
Every number on this page comes from published research. Verify it yourself.
1
SE Ranking, AI Content Experiment (published Mar 2026, updated 2026). 6 human edited AI articles vs 2,000 raw AI articles across 20 domains, tracked with Google Search Console over 16 months. seranking.com/blog/ai-content-experiment
Numbers are reported as published by each source. AI detection has known limits, so classification based figures are directional. Results vary by site, niche, and domain authority. This page summarizes the studies in plain language and is not affiliated with Google, SE Ranking, or Semrush.
