Content

The Evidence of Effort a Reader Can Actually See

Evidence of effort is what a rater sees in minutes. Rank the five signals a reader notices first and cut the four that only look like work.

By David Jubé · Jul 31, 2026 · 13 min read

Table of Contents

Evidence of effort is whatever a stranger can see of your work in the few minutes they give the page, which is a far smaller thing than the work itself.

One hundred and nineteen uses of the word effort sit in Google’s rater guidelines, and none of them measures the hours you spent.

Effort, as the guidelines define it, is the extent to which a human being actively worked to create satisfying content. Raters are told to spend a few minutes examining that content before drawing a conclusion, which is the entire window your work has to show itself in.

Rank what you add by what gets noticed inside that window and you can fix a thin page with one change instead of five.

Key takeaways

Google’s rater guidelines define effort as the extent to which a human being actively worked on the content, and a rater judges it from the page alone.
Rank evidence by visibility, not by what it cost you. A number in a heading gets seen; a careful nuance in paragraph nine does not.
Five things read as proof: an original number, a checkable named example, a visual you made, a position you will defend, and a precise figure in place of a range.
Four things read as work and carry none: padded length, a restated heading, a stock image, and a hedge.
AI can arrange evidence you already hold. It cannot produce the number, the example, or the position, because it was not there.

Effort, as Google’s Raters Are Told to Judge It

Effort is the term Google’s Quality Rater Guidelines use for the extent to which a human being actively worked to create satisfying content. It is one of four things a rater weighs when scoring the main content of a page. The word appears 119 times in the September 2025 edition, and not one of those uses asks how long the work took.

So effort is a fact about the work, judged entirely from the artifact the work leaves behind.

That artifact is your page. It is also the only thing a rater is given, and the only thing a reader was ever going to get.

The guidelines put effort alongside three companions, and the four together decide the quality of a page’s main content:

Effort: the extent to which a human being actively worked to create satisfying content.
Originality: whether the content offers something unique that is not already available elsewhere, and whether this page is the original source.
Talent or skill: whether the content is made well enough to give a visitor a satisfying experience.
Accuracy: for informational pages, whether the content is factually correct, and on health or money topics, consistent with expert consensus.

Raters are also told to spend a few minutes examining the main content before drawing a conclusion about it. Minutes are the budget you write against.

Why Visible Proof Beats Invisible Work

Nothing about your process reaches the page on its own. A founder who compares five suppliers over an afternoon has done the work, and the sentence “suppliers are vetted carefully” carries none of that afternoon forward.

That sentence would read exactly the same if nobody had been vetted at all. Apply the test everywhere: could this line survive if the work behind it had never happened.

Evidence is the part of the work that cannot survive its own absence.

Google’s content and quality questions ask a version of the same thing from the other side. Does the content provide original information, reporting, research or analysis, and does it go beyond the obvious.

Three sentences that fail on sight:

“Quality ingredients, responsibly sourced.”
“Results depend on your situation.”
“Built on years of industry experience.”

Each one would be true of a competitor with the name changed. The first E in E-E-A-T is experience, added in December 2022 so raters could ask whether content was produced with actual use of a product or an actual visit to a place.

Experience and effort get proven with the same class of detail, which is why this article is the Evidence Test half of the pair I set out in what Google actually penalizes about AI content. It is also the half you can act on in one afternoon.

Evidence of Effort, Ranked by What a Reader Notices First

Not every kind of proof gets seen. Nielsen Norman Group’s reading research puts 79% of web users scanning rather than reading, so the larger share of what you add meets an eye moving fast rather than a reader working through it.

Eyetracking of 232 users found the same scan shape holding across many different sites and tasks. Readers sweep twice across the upper part of the content, then run their eyes down its left side.

Proof placed outside that path is proof nobody spends attention on.

So rank evidence by where it gets seen, not by what it cost you to produce:

A number in a heading or an opening line.
A visual you made yourself.
A named example a reader could go and check.
A position stated plainly enough to disagree with.
A precise figure where a range would have been easier.

Seen in the Scan: Numbers and Visuals

Numbers and images survive scanning because neither has to be read in sequence. A figure in a subhead lands inside the horizontal sweep, and a chart you built lands wherever the eye stops.

A number also survives being lifted out of context, which a carefully qualified sentence does not. That property is why it sits at the top of the order rather than second.

Visuals sit next to it for the same reason, on the condition that you made the thing. A chart drawn from your own figures is proof, and a licensed photograph is decoration.

Seen on a First Read: Examples and Positions

A named example and a stated position each cost a reader one sentence of attention, which puts them a tier down from a number. Both take a minute to add and neither can be invented.

A position is also the first thing a nervous draft hedges away. Copyblogger’s case for writing with authority is that readers are tired of fluff and hedging, and that a stated view is the one thing a summary can never contain.

Andy Crestodina makes the harder version of the same point: strong opinion and original research are the two content types a model cannot make, because everything it produces is an average of what it has already read.

Named examples work for a different reason: if a reader could look it up and find you were right, the sentence carries weight even when nobody looks.

Seen Only on a Careful Read: Precision

Precision is the tier a scanning reader may never register and a careful one always does. Swapping “about a cup” for a measured figure changes nothing visually and changes what the sentence is worth.

It is also the least work of the five, because the precise number is usually sitting in your notes already.

Rank order matters because a founder normally has time to add one thing, not five. Add the one that gets seen.

Founder Insight: One Specific Beats Five Vague Adding all five signals to a weak paragraph usually produces a longer weak paragraph. Pick the one a scanning reader hits first, put it in, and leave the rest of the paragraph alone.

Four Things That Look Like Work and Carry None of It

Cutting is the faster half of this. Four habits read as effort at a glance, cost real words, and leave the page exactly as anonymous as it was before:

Padded length: a paragraph stretched from three sentences to eight without one new fact entering it.
The restated heading: an opening sentence that says the heading again in longer words before the section starts.
Stock imagery: a licensed photograph anyone can buy, signalling the opposite of what it was chosen to signal.
The hedge: “Results may vary” and “generally speaking” are sentences declining to make a claim a reader could test.

Length is the habit people defend hardest. Backlinko’s roundup of search statistics puts more than 66% of pages at zero backlinks, citing Ahrefs, which is a fair picture of how much published writing nobody ever points at.

Short or long, the default outcome for a published page is that nothing points at it. Length belongs to the intent rather than to a word target, which is the same argument arriving from the other direction.

The Nine, Side by Side

Each row names the tier from the ranking above, so the two lists can be read as one. The guideline terms in the verdict column are the four quoted earlier from the rater guidelines.

Signal	Where a reader meets it	Verdict
Number in a heading or opening line	In the scan	Counts, under effort
Visual you made	In the scan	Counts, under originality
Named, checkable example	On a first read	Counts, under effort
Stated position	On a first read	Counts, under talent or skill
Precise figure in place of a range	On a careful read	Counts, under accuracy
Padded length	Everywhere, doing nothing	Looks like work
Restated heading	First line of a section	Looks like work
Stock image	In the scan, then dismissed	Looks like work
Hedge	Wherever a claim was due	Looks like work

Cut the bottom four and the page gets shorter. That is the correct outcome, and it is why cutting is worth doing before adding.

Where AI Helps Assemble Evidence and Where It Cannot Supply It

AI is good at arranging evidence and incapable of producing it. That is not a criticism of the tools, it is a description of what they have access to.

80% of marketers now use AI for content creation, which means fluent, well-organised prose has stopped separating one page from another.

The audience is not neutral about this either. Pew found 51% of the US public more concerned than excited about AI in daily life, against 15% of the AI experts surveyed.

The split in the work is clean once you see it:

Turning your notes from an afternoon of supplier calls into readable prose.
Formatting numbers you already hold into a table a reader can scan.
Drafting the structure so your time goes to the specifics.
Supplying the number is out of reach, because the model was not in the room.
Supplying the position is out of reach, because it has no stake in being wrong.

The wider division of labour sits in use AI to write without sounding like AI, and the sentence habits that undo good evidence are in the AI tells that cost you readers, not rankings.

One rule holds across every tool. If a sentence would still be true with your specifics swapped for a competitor’s, nothing was added by writing it.

Rewriting One Weak Paragraph, Step by Step

Take a hypothetical two-person studio selling hand-thrown ceramics, with a page about choosing a glaze. Here is the sort of paragraph a first draft produces, and none of it is wrong.

“Choosing the right glaze matters for the finished piece. Colour, food safety and how the glaze behaves in the kiln all contribute to a good result. It is important to do your research before committing.”

Nothing there is false. Nothing there is evidence either, and those three sentences would sit unchanged on any ceramics site anywhere.

Three moves fix it:

Replace one general claim with a figure the studio already holds.
Name one thing a reader could go and verify.
State the position the paragraph has been avoiding.

Applied, the same hypothetical paragraph becomes something closer to this.

“Two of the three glazes in this range are food safe at the temperature the kiln actually holds, and the third is not, which is why it is sold for vases only. Test tiles are fired in the same load as the pieces, so the colour on the tile is the colour you get.”

Same length. The second version could not have been written about a different studio, and here that is the only difference that counts.

A definition written to be quoted verbatim is this same discipline applied to a single sentence.

The move is not “add detail”. The move is to add the detail that would be false if it were copied onto someone else’s page.

What to Check Before the Next Draft Publishes

Read the draft once for this and nothing else. Grammar and fact-checking are separate passes, and mixing them means the evidence question quietly gets skipped.

One question per paragraph: could this be true, word for word, on a competitor’s page with the names changed.
Mark the failures, do not fix them yet: a whole-draft view tells you which sections are hollow and which are merely plain.
Add one signal per flagged paragraph: take it from the ranked five, starting at the top.
Put the strongest specific where the scan goes: a number in paragraph nine is a number nobody meets.

Formatting decides whether any of this survives contact with a reader. Comparison tables and formats LLMs love to cite and content that ranks and gets cited by AI both cover the layout side of the same problem.

Writing for E-E-A-T is the wider version of this argument, taking in trust and authority as well as effort.

This check tells you nothing about where a page will rank. It tells you whether the page holds anything a competitor’s page does not, which is a smaller question, and the one worth settling first.

Frequently Asked Questions

What counts as evidence of effort on a page?

Evidence of effort is anything on the page a stranger can point to as proof of work. An original number, a named example a reader could verify, a visual you made, a stated position, or a precise figure in place of a range. Raters judge what shows.

Does original data mean I need a research team?

No. Original data means numbers you already hold that nobody else can publish: your own order counts, turnaround times, test results, or before-and-after figures. One number pulled from your own records is original data. Scale is not the thing that makes it count as evidence.

How specific does an example have to be to count?

Specific enough that a reader could go and check it. Name the product, the project or the number precisely enough that being wrong would be visible to anyone who looked. Anonymous stand-ins prove nothing at all, because there is no way for a reader to verify them.

Do I need custom visuals, or will stock images do?

Stock images do not count. A visual counts when you made it: a photograph you took, a chart built from your own numbers, a screenshot of your own dashboard. Orbit Media’s survey found bloggers who add a lot of visuals report strong results far more often.

Can a rater tell how much work went into a page?

Only from the page. Raters are told to spend a few minutes examining the main content and judge what sits in front of them, so a week of research and a hurried afternoon look identical unless the research left something visible behind on the page.

Which single change adds the most evidence to a weak draft?

Pick one paragraph and add a single specific a competitor could not copy: a number from your own records, a checkable example, or a plain statement of your position. One specific placed where a scanning reader meets it beats five buried further down.

Continue Reading

More From This Series

More from TDM Insights

Explore TDM Insights Topics