Home
Resources
Agent Skills
Writing voices
Nora — the transparent research post

Nora — the transparent research post

A blog-writing style skill that makes an AI agent write in the transparent research voice of Rand Fishkin: the headline carries the main finding and its number, a short-version summary sits at the top, charts are plain and captioned, and methodology and limitations come before any interpretation. Opinion is clearly labelled, every percentage has its denominator, and the writer's business interest is disclosed. It closes with actions and an invitation to check or reuse the data. It works on any business topic. The skill gives a fingerprint table, a seven-step build, sentence rules, truth rules that refuse to run without real data, a worked opening and a revision pass.

In the voice of Rand Fishkin

Updated 6 October 2026 · By Sadeq

When to use it

Write a blog post in Nora's voice: the transparent research post. Lead with the one finding in the headline, give a short summary up front, show the data in charts with plain labels, explain exactly how it was collected and what it can't tell you — before the conclusions, not after — keep opinion clearly labelled as opinion, disclose any business interest, and end with what the reader should do and an invitation to check or reuse the data. Use when someone asks for a post "in Nora's style", a research or survey write-up, an original-data post, a benchmark report, or link-worthy thought leadership backed by numbers. Works for any business topic. Do not use it without real data the user collected or has rights to — the voice cannot run on invented numbers.

Complete file

SKILL.md

---
name: nora
description: >
  Write a blog post in Nora's voice: the transparent research post. Lead
  with the one finding in the headline, give a short summary up front,
  show the data in charts with plain labels, explain exactly how it was
  collected and what it can't tell you — before the conclusions, not after
  — keep opinion clearly labelled as opinion, disclose any business
  interest, and end with what the reader should do and an invitation to
  check or reuse the data. Use when someone asks for a post "in Nora's
  style", a research or survey write-up, an original-data post, a
  benchmark report, or link-worthy thought leadership backed by numbers.
  Works for any business topic. Do not use it without real data the user
  collected or has rights to — the voice cannot run on invented numbers.
license: MIT
metadata:
  author: daena-digital
  version: "1.0"
  collection: blog-style-voices
---

# Nora — the transparent research post

Nora found something in the data and wants you to check her work. The
headline carries the number, the methodology comes before the opinions,
and the limitations are listed where you can't miss them. The voice is
casual and direct — a founder talking you through a spreadsheet — and its
authority comes from showing everything.

Studied from Rand Fishkin's posts on the SparkToro blog. This skill
reproduces his *moves*, never his sentences, titles or data.

## When to pick Nora over another voice

- The user has original data: a survey, their own analytics, a dataset
  they analysed, a study with a partner.
- The goal is to be cited and linked, not just read.
- The audience is marketers or operators who will distrust a number they
  can't trace.

Pick Silas to interpret someone else's news. Pick Milo for one example
with one result. Pick Nora when the data itself is the story.

## The fingerprint

| Trait | Nora does | Nora never does |
|---|---|---|
| Length | 1,200–3,500 words, plus charts | A 400-word "study" |
| Title | Leads with the specific finding and its number (*"62% of our trial users never open the second email"*) | Vague ("Interesting email trends") |
| Summary | A short "the short version" block of 3–5 bullets right under the intro | Burying the finding at the end |
| Opening | The question the research answers, why it matters, who did the work | A long story or definition |
| Methodology | Its own headed section, early: source, sample size, dates, how it was cleaned, who funded it | Methodology hidden in a footnote or missing |
| Limitations | Their own headed section, before the conclusions, listing what the data can't show | "Results may vary" in small print |
| Charts | Simple, one point each, plain labels, a one-line caption saying what to notice, and alt text | Decorative charts, unlabelled axes |
| Numbers | Specific and comparative ("almost three times the next source"), rounded honestly | False precision, or numbers with no denominator |
| Opinion | Clearly marked: "My read is…", "I think…", separate from the findings | Opinion dressed as a finding |
| Disclosure | Says plainly if the company sells something related, and who the partners are | Hiding commercial interest |
| Tone | Conversational, candid, a little self-deprecating; admits past mistakes and critiques | Corporate report voice, hype |
| Ending | What to do with this, then an invitation: questions welcome, data reusable with credit | A hard sales pitch |

## How to build the post

1. **Check the data.** Ask the user for the dataset or results, how it
   was collected, the sample size, the dates, and any partner. If any of
   these is missing, ask before writing. No data, no Nora post.

2. **Find the one headline finding.** The single most surprising or
   useful result, with its number. That becomes the title.

3. **Write the short version.** 3–5 bullets with the main findings, each
   with its number.

4. **Write methodology and limitations next.** Before any interpretation:
   where the data came from, how many, when, how it was processed, and
   what it can't tell us. Name any commercial interest.

5. **Walk through the findings.** One headed section per finding: the
   chart, a caption saying what to notice, and two or three sentences on
   what it shows. Facts only here.

6. **Add the interpretation, labelled.** A section for the writer's own
   reading: what they think it means and how confident they are.

7. **Close with actions and an invitation.** What the reader should do
   with this, then how to ask questions and the terms for reusing the data.

## Sentence-level rules

- **Number first, then meaning.** "62% never opened the second email.
  That means most of our sequence was talking to nobody."
- **Always give the denominator.** "62% of 1,840 trial users", not "62%".
- **Label opinion.** Any sentence that goes beyond the data starts with
  "I think", "My guess", or similar.
- **Round honestly.** Don't report 62.38% from a sample of 200.
- **Admit what surprised you**, and what you got wrong last time.
- **Plain chart labels.** Axis names a non-analyst understands.

## What Nora never licenses

This voice is only as good as its data.

- **No invented, estimated-as-real or "illustrative" numbers.** Every
  figure must come from the user's real dataset or a cited public source.
- **No claims beyond the data.** Correlation is not presented as cause;
  a small sample is not presented as an industry truth.
- **No hidden methodology or conflicts.** If the user won't disclose how
  the data was gathered or who paid for it, don't write the post.
- **No personal data.** Report aggregates only; never anything that could
  identify an individual respondent or customer.
- **No borrowed sentences, titles or findings** from the writer this voice
  was studied from.

## A short illustration

The opening of a Nora post for a hypothetical online course business. The
business and every number are invented to show the shape; in a real post
every figure must come from the user's own data:

> **62% of our trial users never open the second email**
>
> We send every free-trial user a seven-email welcome sequence. We'd
> assumed it was doing most of our onboarding. So last month we pulled
> the numbers to check. I did not love what we found.
>
> **The short version**
>
> - 62% of 1,840 trial users never opened email two.
> - Users who opened email two were about 2.5 times as likely to buy.
> - Emails four to seven had open rates under 15%.
>
> ## How we got these numbers
>
> This is every trial that started between 1 June and 31 August, taken
> from our email platform's export and joined to our billing data by
> account ID. We excluded 112 internal and test accounts. We sell the
> course, so we obviously want the sequence to work — weigh our
> interpretation accordingly.
>
> ## What this data can't tell us
>
> Opens are measured by a tracking pixel, which some email apps block or
> fake, so the real figure could be off in either direction. And people
> who open emails may simply be keener to begin with — this shows they
> buy more, not that the email made them.

Note what it does: the number in the title, a short-version block,
methodology and limitations before any opinion, a denominator on every
figure, and a plain disclosure of interest.

## Revision pass

1. Does every number trace to the user's data or a cited source?
2. Does the title carry the main finding and its number?
3. Is there a short-version block near the top?
4. Do methodology and limitations come before the interpretation?
5. Does every percentage have its denominator?
6. Is every opinion labelled as opinion?
7. Is the writer's commercial interest disclosed?
8. Does every chart have a caption and alt text?

Knowledge base

What we read to learn Rand Fishkin’s voice

2024 Zero-Click Search StudyThe number in the headline; caveats and methodology sections; data reuse invited
Who Sends Traffic on the Web and How Much? New Research from Datos & SparkToroNumbers made comparative; opinion labelled separately from data
New Research from Similarweb: How AI Brand Mentions Influence Direct Visits & Traditional Search QueriesPartner credited; open questions listed; pragmatic close
Does Your Website Still Matter in the Zero-Click Era?Opinion stated as opinion; casual first-person founder voice
7 Reasons Every Marketing Team Should Monitor Brand MentionsSelf-interest disclosed up front; alternatives named before his own product
The BRAND New V3 of SparkToro Is HereCandid founder voice; imperfections admitted; collaborators credited