Table of contents
Introduction
English version of an article first published in Polish on 4 October 2026. Read the Polish original.
On 1 October 2026 Google added to its documentation for creators what until now had been described mainly in the handbook for the people who rate the quality of search results. There is a new definition of the main content of a page and four characteristics by which that content is assessed: effort, originality, talent or skill, and accuracy.
The algorithm has not changed because of this. What has changed is what Google says plainly to the people who write pages. Instead of a list of self-assessment questions there is a named set of criteria, with examples of what Google considers a lot of work and what it considers none. One of the examples of a lack of effort is AI text published in bulk and with no oversight.
I compared the versions of both pages from before and after 1 October, sentence by sentence, and read the relevant parts of the rater guidelines. That makes it clear what is really new, what was already there and what does not appear in the official changelog at all.
What you will learn:
- What exactly changed on two pages of Google’s documentation on 1 October 2026, based on a comparison of the versions before and after the change
- How Google’s new definition of main content reads, what Google counts as the main content of a page and why tabs, reviews and tools count too
- How Google defines effort, originality, talent or skill, and accuracy, and which examples it gives
- Whether these characteristics are a ranking factor and whether they are connected to the ongoing September spam update
- What changes for content written with the help of AI and for sites with invented authors
- What to check on a business website, in an online shop, on a health or finance site and in content produced at scale
What exactly Google changed
Two pages of the Google Search Central documentation changed. Both carry the same last-updated date, 1 October 2026, and both had previously been updated on 10 December 2025. I compared their content with the versions saved in the Internet Archive on 27 and 28 September. Here is what was added.
What was added to the documentation on 1 October 2026
Page: Helpful, reliable, people-first content
not in the changelogA definition of main content and its five types. Four characteristics assessed by quality raters, with examples. A sentence saying that citing a source does not replace your own effort. A warning against made-up author profiles.
Page: Generative AI content
in the changelogA paragraph saying that models predict the next words rather than retrieve facts, so they can be wrong. A requirement to review all AI content manually before publication, including titles, meta descriptions, structured data and alt text.
The changelog tells half the story
Google keeps a public changelog of its documentation. Under 1 October it has one entry: the guide on using generative AI content was updated “with information from the Search Quality Raters guidelines”. The reason Google gives is “to get our documentation in sync with our presentations we use at our developer events”.
Except that the material from the rater guidelines went mainly to the other page, the one on helpful content. That is where the definition of main content, the four characteristics and the warning against made-up authors are. The AI page gained one paragraph on hallucinations and fact-checking. Anyone who reads only the changelog will miss the most important part of the update.
The explanation about developer events says something too. It looks as if Google is not announcing a new policy but writing down in the documentation what its staff already say at conferences. So the direction was known; what is new is its official form.
What was already there
It is easy to credit the October change with things that were in the documentation before, especially as the changelog mentions “information from the Search Quality Raters guidelines” precisely in connection with the AI page. On the AI page, the reference to sections 4.6.5 and 4.6.6 of the rater guidelines, the recommendation to tell readers how content was created, and the Merchant Center requirements for AI product images and descriptions were already in the December 2025 version. The page on helpful content has long had the “Who, How, and Why” questions about content and the recommendation to add bylines. What is new is what I show in the table above.
The Polish version of the documentation does not have it yet
If you open these pages in Polish, you will find none of what I describe here. So far Google has updated only the English version. The Polish translations are older and contain none of the October changes.
English version and Polish version (as of 4 October 2026)
- Helpful, reliable content: English version of 1 October 2026, Polish version of 18 December 2025. The Polish one has no definition of main content, no four characteristics and no warning against made-up authors.
- Generative AI content: English version of 1 October 2026, Polish version of 31 December 2025. The Polish one has no paragraph on hallucinations and manual fact-checking.
That is why every quotation here comes from the English version. In the Polish original of this article the names of the characteristics are my own translation, given alongside the English terms, because those are the terms under which the change is discussed in the industry. It is also worth knowing that Google’s official Polish translations call the raters “testerzy jakości wyszukiwarki” (search quality testers).
Main content: Google’s new definition
The new section opens with a sentence worth reading twice. According to Google, the rater guidelines stress that one of the most important factors in assessing the quality of a page is the quality of its main content, specifically how effectively it allows the page to achieve its purpose and gives the visitor a satisfying experience.
The definition reads: “Google Search defines "main content" as any part of the web page that directly helps the page achieve its purpose”. In other words, main content is every part of a page that directly helps it do its job. In the rater guidelines the same term is abbreviated to MC, and Google’s older Polish translations render it as “główne treści”. Depending on what the page is for, it can include:
- Primary text and media: the main text, articles, images, audio or video files.
- Interactive features: calculators, online tools, games, search functionality on the page.
- User-generated contributions: forum discussions, customer reviews, comments, user-uploaded media, when they directly fulfil the purpose of the page.
- Tabbed or expanded sections: for example product specifications, safety notes or user reviews.
- Page titles and headings: the visible title and headings that summarise the topic and help people decide whether to visit the page.
A product page through a rater’s eyes
The rater guidelines divide every page into three parts. Besides the main content there is supplementary content, which improves the experience but does not directly fulfil the purpose of the page, for example navigation or links to other products. The third part is advertising and other forms of monetisation. The guidelines say plainly that the mere presence of ads is not a reason for a low rating.
What follows in practice
Shops gain the most from this definition. For years the question kept coming back whether content in tabs “counts” for Google. The documentation now lists it explicitly as part of the main content, with examples: specifications, safety notes, reviews. The rater guidelines add a caveat that the content of a tab can also be supplementary content or advertising, depending on what is in it. A tab with specifications is main content, but a tab with a partner’s offer is not.
The second thing is tools. A calculator, a configurator or a map is not an add-on to the text but main content in its own right. For many service websites that is good news, because a well-made tool often answers the question better than another two thousand words.
The third is titles and headings. Google counts them as main content and describes their job: to summarise the topic and help people decide whether to visit the page. The self-assessment questions on the same page have long warned against exaggerating or shocking titles. Now it is clear that the title is assessed together with the content, not alongside it.
One caveat that gets lost in summaries. The text speaks of “the visible title and headings throughout a page”, which in practice means the H1 and subheadings. It does not refer directly to the <title> element, which is what appears in search results. Google covers that one separately, on the page about AI content, together with the meta description and structured data. If it was produced with the help of AI, it has to go through the same manual review as the rest of the content.
The four characteristics
Google then lists four fundamental attributes of main content that quality raters are trained to evaluate. The descriptions on the page for creators are close to those in section 3.2 of the rater guidelines, but not identical. Google added its own examples, which in practice are the most valuable part of the change.
How raters measure main content
Effort
human workHow much human work went into the content or into the system that creates it.
High: Original analysis, a poem translated by hand, an interactive map built for a game.
Low: Pages generated automatically from feeds, large amounts of AI text with no oversight or editing.
Originality
something newHow far the content offers information or a perspective that other websites do not have.
High: Your own data, tests, photos, project experience, conclusions nobody else has reached.
Low: Rewriting or paraphrasing other people’s texts, even with the source cited.
Talent or skill
done wellWhether the content shows the skill needed to give the visitor a satisfying experience.
High: Clearly written text, well-produced video, a tool that works.
Low: Sloppy text, chaotic material, broken features. Note: not all content requires an expert.
Accuracy
factsFactual correctness on informational pages, and for YMYL topics high accuracy and consistency with expert consensus.
High: Facts checked at the source, health or financial advice in line with what specialists currently know.
Low: Easy-to-spot errors, outdated data, advice that contradicts the consensus.
Effort
Google defines effort as the extent to which human work went into creating the content or the systems powering it. The second half matters. Effort is not only writing a text but also building a tool, a database or a carefully worked-out template. Google’s examples of high effort are writing original analysis, translating a poem manually and building a custom interactive map for a game.
The examples of a lack of effort are just as specific: automatically generating pages from feeds, and using generative AI to produce large amounts of text “without manual oversight or curation”. Note that qualification. Google does not write that AI text means no effort. It writes that the lack of effort lies in mass production nobody checks.
One sentence in this point is easy to overlook, and it concerns a large part of the internet: “Attribution or giving credit to other sources doesn’t replace the need for original effort.” An article that faithfully summarises three pieces from abroad and links to each of them can still be content with no contribution of its own. The rater guidelines say the same more sharply: a page whose main content is almost entirely copied or paraphrased, with no effort and no added value, gets the lowest rating even if it credits the author of the original.
For Polish online shops, one example from section 3.2 of the guidelines matters. A site that offers users machine translation as a service has effort in it, because somebody built that system. But automatically generating thousands of pages by running freely available content through an off-the-shelf translator, with no oversight or editing, is not effort. The example in the guidelines is about other people’s content, but a shop that translates its whole catalogue with one button and puts it on a new market unchecked is very close to that description.
Originality
Originality, according to Google, is the extent to which content offers unique, original information or perspectives that aren’t already available on other websites. The rater guidelines add a helper question: if other sites have similar content, is this page its original source?
This criterion is the hardest to meet in how-to articles, because most topics have already been covered. Originality does not have to mean a new topic, though. It can be your own example from work, the result of a test, a photo you took yourself, data from a project or a conclusion nobody else has drawn. This article is a good example: dozens of sites are writing about the same change today, but a sentence-by-sentence comparison of the page versions is something I did not find in the coverage I read.
Talent or skill
The third characteristic is talent or skill: whether the content shows the competence needed to give visitors a satisfying experience. Google gives examples: clear writing, well-produced video, functional page tools.
The qualification Google added straight away matters: not all content requires specialised expertise. The example in the documentation is a person sharing how they found a new way to clear their driveway of snow. You do not need to be an expert for that. This is good news for small businesses and enthusiasts. Skill here means above all careful execution, not an academic title.
Accuracy
The fourth characteristic has two thresholds. Informational pages should be factually accurate. For topics that could significantly affect people’s lives or well-being, the YMYL category (from “Your Money or Your Life”), the content must be highly accurate and consistent with established expert consensus.
That second part has practical consequences. An article on health or finance may contain not a single error and still be rated poorly if it presents as certain a view most specialists disagree with. In the self-assessment questions Google has also long asked about “easily-verified factual errors”. So accuracy starts with simple things: dates, numbers, names and quotations.
Is it a ranking factor
The honest answer is: not directly. The four characteristics describe how people assess content, not how an algorithm scores it. On the same page Google writes that raters have no control over how pages rank and rater data is not used directly in its ranking algorithms. It compares the ratings to the feedback cards a restaurant collects from diners. On the AI page it adds that the rater guidelines “are not a guide to ranking first in Google”.
That does not mean the characteristics are irrelevant. Google uses human ratings to check whether changes to its algorithms improve the results. If raters are trained to give the lowest rating to content with no effort, an algorithm change that lifts such pages should do badly in testing. Algorithms do not measure effort the way a person does, but they are developed towards results that people will rate highly.
What about the contentEffort attribute from the leak?
The 2024 leak of Google’s internal API documentation contains a contentEffort attribute in a module with page quality signals, described as an effort estimate made by a language model for article pages. It is tempting to connect it with the new documentation. It is worth remembering, though, that the leak shows the state of things in 2024, says nothing about the weight of the attribute and does not prove that it is used today. It shows only that in 2024 Google had in its systems an attribute described as a measurement of effort.
The September spam update in the background
The documentation changed while an update was in progress. According to the Google Search Status Dashboard, the September spam update started on 24 September 2026 and is still rolling out. That is longer than the two previous updates of this type: the August one took 2 days and 16 hours, the June one 2 days and 1 hour.
Articles are already appearing online claiming that this update “hit AI sites”. There is no evidence of that. The sites I build with the help of AI myself are growing at the moment. Google has not announced which practices the update targets, and the coincidence of dates with the new documentation settles nothing. If you have seen drops in recent days, first check whether they affect pages that fit the description of scaled content abuse before you put them down to anything else.
Where this came from
The October change is the latest step in a process that has been going on for four years. First Google treated content written for search engines as a separate problem, then it built that assessment into the core algorithm, and finally it described the criteria, first for raters and now for everyone.
From helpful content to the four characteristics
-
August 2022
The first helpful content update
a separate system that lowers the assessment of sites with content written for search engines
-
March 2024
The system becomes part of the core algorithm
the end of separate helpful content updates; new spam policies, including scaled content abuse
-
January 2025
Sections 4.6.5 and 4.6.6 in the rater guidelines
scaled content abuse, and content with no effort, originality or added value; the first definition of generative AI
-
11 September 2025
The current version of the guidelines
the source of the description of the four characteristics of main content
-
24 September 2026
The September spam update starts
still rolling out according to Google’s status dashboard
-
1 October 2026
The four characteristics in the documentation for creators
plus the definition of main content, the warning against made-up authors and manual verification of AI content
The most important point in this history is January 2025. That is when Google rebuilt the chapter on spammy pages in the rater guidelines. Section 4.6.5 on scaled content abuse appeared, along with section 4.6.6 on main content created with no effort, originality or added value. According to it, the lowest rating goes to a page on which all or almost all of the main content is copied, paraphrased, embedded, auto-generated or AI-generated, or reposted from other sources with no effort, originality or value for visitors.
The same section contains a sentence worth knowing by heart, because it closes the “AI or human” argument: the use of generative AI tools does not by itself determine the level of effort or the quality rating of a page. The guidelines add that such tools can be used to create very good content as well as the worst. What is assessed is the result and the work put in, not the tool.
AI content after the change
The page on AI-generated content gained one paragraph, but a very direct one. Google reminds readers that generative models “don’t retrieve facts, but predict a likely sequence of words” based on their training data, so their output may contain inaccuracies known as hallucinations. It goes on to say that it is “critical” to fact-check manually and review all AI-generated content for accuracy and trustworthiness before publishing.
This requirement also covers elements that hardly anyone reads before publication but that can appear in search results: page titles, meta descriptions, structured data and image alt text. That matters especially for shops and sites that generate these elements in bulk. A meta description with an invented price or structured data with a rating that does not exist is exactly the kind of error nobody notices until a customer sees it in the results.
What stayed the same
The rest of the AI page was already there, and it is worth knowing in full, because it is the shortest official instruction on using AI in content:
- Generative AI can be useful for research and for adding structure to original content. Generating many pages with no value for users, on the other hand, may violate the policy on scaled content abuse.
- Google recommends telling readers how content was created where that makes sense, for example by explaining what automation was used for.
- In Merchant Center, AI-generated images must carry the IPTC
DigitalSourceTypemetadata with the valueTrainedAlgorithmicMedia. - AI-generated product data, such as the title and description, has to be specified separately in Merchant Center and labelled as AI-generated.
If you use AI when writing, I have described how to use AI in SEO sensibly, and when covering Anthropic’s watermark I wrote about what AI content detection does not prove. The new documentation fits both pieces well: Google does not ask whether a model wrote the content, only whether somebody checked it and whether it adds anything new.
What to check on your site
The documentation is the same for everyone, but the conclusions depend on the type of site. I have set them out for the four most common situations. If your site combines several of them, for example a shop with a blog, go through each list that applies.
01. Business website and service blog
- Sign your articles with a real author and link to a page that shows their experience.
- For every article, ask yourself what you add beyond what is already in the top ten: your own examples, project data, photos, conclusions.
- A language model can help with research and proofreading, but you check facts, numbers and dates at the source.
02. Online shop
- Do not paste in manufacturer descriptions. That is the same content dozens of competitors have.
- Specifications, reviews and safety information in tabs are main content. They should be complete and available without logging in.
- AI images in Merchant Center must carry IPTC metadata, and generated titles and descriptions in the feed have to be labelled.
- The guidelines give bulk machine translation with no oversight as an example of a lack of effort. Translating a whole catalogue with one button comes very close to it.
03. Health, finance, law (YMYL)
- The content has to be consistent with expert consensus, not merely “free of errors”.
- Have a specialist review the text on its merits and credit that review by name.
- Change the update date only when the content has really changed. Google lists shifting dates as a warning sign.
04. Content produced at scale
- Check whether the page would make sense to someone who arrives on it directly rather than from a search engine.
- Every page built from a template needs data the others do not have. Hundreds of pages with no content of their own fit the description of scaled content abuse in the guidelines.
- Introduce editorial oversight: who reads, who corrects, who decides on publication.
The common denominator of all four lists is the question Google calls, on the same page, “perhaps the most important”: why is this content being created? If the answer is “to help people who arrive on it directly”, you are on the right side. If it is “to catch search engine traffic”, none of the four characteristics will save it. For expert content, a structure in which the answer comes first also helps. I described it when writing about the BLUF method in SEO.
My view
On the four characteristics, Google has said nothing that was not already in the rater guidelines. And yet I consider this change important, for a simple reason. Until now, the argument “that is how the rater guidelines put it” sounded, in a conversation with a client, like quoting an internal handbook that Google itself says is not a guide to ranking. Now the same criteria stand in the documentation for creators, next to the self-assessment questions. They are easier to talk about and harder to brush aside.
What I like most is how Google described effort. Work put into the system that creates content is effort too. That is a fair way of putting it: a good configurator, a well-thought-out database or a tool that really solves a problem has value even if there is not a single hand-written paragraph next to it. The line does not run between human and machine, but between work and the lack of it.
On this blog, every article ends with a note on how it was made and what a language model helped me with. I added it in July, because I think the reader has a right to know. Google’s documentation calls this the “How” question and recommends exactly this practice.
AI is not a problem for Google. The problem is a text nobody read before publication and an author who does not exist. Both were bad practice long before language models. Now Google has simply called it by its name in documentation anyone can read.
FAQ: common questions
What did Google change in its content guidance in October 2026?
On 1 October 2026 Google updated two pages of the Search Central documentation. The page on helpful, reliable content gained a definition of main content, a description of four characteristics assessed by the quality raters Google employs (effort, originality, talent or skill, and accuracy) and a warning against made-up author profiles. The page on AI-generated content gained a paragraph on model hallucinations and on the need to fact-check manually before publishing.
How does Google’s new definition of main content read?
Since 1 October 2026 Google’s documentation says: “Google Search defines "main content" as any part of the web page that directly helps the page achieve its purpose”. Google includes text and media, interactive features (calculators, tools, search functions), user-generated contributions when they serve the purpose of the page, the content of tabs and expandable sections, and the visible title and headings. In the rater guidelines the term is abbreviated to MC.
Are the new passages in the Polish version of Google’s documentation?
As of 4 October 2026, no. Google updated only the English version. The Polish version of the page on helpful content dates from 18 December 2025 and the page on AI content from 31 December 2025. They contain no definition of main content, no four characteristics, no warning against made-up authors and no paragraph on hallucinations. To read the current text you have to switch the page to English.
Is the effort put into content a ranking factor?
Google does not say so. The four characteristics describe how people assess content, and according to Google their ratings do not affect rankings directly; they only help check whether the algorithms work well. The 2024 leak of Google documentation contains a contentEffort attribute, described as an effort estimate made by a language model, but it is not known whether it is used today or how much weight it carries.
Does Google penalise content written by AI?
Not for the use of AI as such. The rater guidelines say plainly that using generative tools does not by itself determine the level of effort or the rating of a page. The lowest rating goes to content that is copied, paraphrased or generated with no effort, originality or added value, whether a person or a model did it. Generating pages at scale with no value for users can, however, violate Google’s spam policies.
Does content hidden in tabs count for Google?
According to the new definition, yes, if it serves the purpose of the page. Google explicitly lists information behind tabs, such as product specifications, safety notes or user reviews. The rater guidelines add, however, that the content of tabs can also be supplementary content or advertising, depending on what is in them.
Does AI content have to be labelled?
Google does not require it in search results. It recommends telling readers how the content was created where someone might ask. A requirement applies in Merchant Center: AI-generated product images must carry the IPTC DigitalSourceType metadata with the value TrainedAlgorithmicMedia, and generated titles and descriptions have to be specified separately and labelled as AI-generated.
Is a made-up author with an active social media profile credible to Google?
According to the documentation, no. Google describes fabricating creator profiles, meaning made-up names, generated headshots and false credentials meant to make content look as if it was written by human experts, as deception. A maintained social media profile does not change the fact that the author does not exist. Google does not reveal how it detects such personas, but the guidelines tell raters to check the creator’s reputation beyond the site, and a persona lacks the traces a real specialist leaves. For sponsored articles signed by a fictional expert there is an additional legal risk in Poland, linked to the ban on undisclosed advertising.
What does the abbreviation EOT/SA mean?
It is shorthand from industry coverage for the four characteristics: Effort, Originality, Talent or Skill, and Accuracy. Google does not use it, neither in the documentation nor in the rater guidelines. It is better to talk about the characteristics themselves, because the abbreviation suggests an official term that does not exist.
Is the September spam update connected to the new documentation?
There is no evidence of that. The spam update started on 24 September 2026 and, according to Google’s status dashboard, is still rolling out. The documentation was changed on 1 October. The coincidence of dates proves nothing, and Google has not announced which practices the update targets.
Worth remembering
- On 1 October 2026 Google added to its page on helpful content a definition of main content and four characteristics that quality raters use to assess a page: effort, originality, talent or skill, and accuracy.
- The documentation changelog lists only the update to the page on AI content. The new sections on main content and the warning against made-up authors arrived without an entry of their own.
- Google also counts tools, user reviews, the content of tabs, and titles and headings as main content, as long as they serve the purpose of the page.
- Google gives pages generated automatically from feeds and bulk AI text with no oversight as examples of a lack of effort. Citing the source does not replace your own contribution.
- Human ratings do not affect rankings directly. They are used to check whether the algorithms work the way Google wants, so they show the direction in which those algorithms are being developed.
- AI content has to be checked manually before publication, including titles, meta descriptions, structured data and image alt text.
- Google calls made-up author profiles, generated headshots and false credentials deception and a signal of a low-quality page.
Summary
Since 1 October 2026 Google has been telling creators plainly how it judges content: by the effort, originality, talent or skill, and accuracy of the main content. It also counts tools, reviews, tabs and headings as main content. AI content requires manual verification, and made-up authors are deception in Google’s eyes.
None of these things is a new ranking factor and none changes a site overnight. Reviewing your own content against the four characteristics is a good exercise all the same. The first conclusions usually come quickly, because it is easy to point to articles rewritten from other sources or an author who does not exist. Fixing it is harder. Adding original data, examples and verified facts to dozens of pages is work for weeks, not for one afternoon.
Sources and materials
Google’s documentation, archived versions of the pages, the rater guidelines and industry coverage on which I base the facts, quotations and dates given here:
- Google Search Central – Creating helpful, reliable, people-first content (English version, updated 1 October 2026; the Polish version of 18 December 2025 does not contain the changes)
- Google Search Central – Google Search’s guidance on using generative AI content on your website (English version, updated 1 October 2026; the Polish version of 31 December 2025 does not contain the changes)
- Google Search Central – Latest Google Search Documentation Updates (documentation changelog, English version)
- Internet Archive – the helpful content page as captured on 28 September 2026
- Internet Archive – the AI content page as captured on 27 September 2026
- Google – Search Quality Rater Guidelines, General Guidelines (version of 11 September 2025)
- Google Search Status Dashboard – ranking update history
- Google Search Central Blog – Google Search’s guidance about AI-generated content (February 2023)
- Google Search Central Blog – New ways we’re tackling spammy, low-quality content on Search (March 2024)
- Pomoc Facebooka – Jakie imiona i nazwiska są dozwolone na Facebooku (Facebook Help Centre, in Polish: which names are allowed on Facebook)
- Meta Transparency Center – Authentic identity representation (Community Standards)
- TikTok – Community Guidelines: Integrity and Authenticity
- Ustawa z dnia 23 sierpnia 2007 r. o przeciwdziałaniu nieuczciwym praktykom rynkowym, art. 7 pkt 11 (Polish Act of 23 August 2007 on counteracting unfair market practices, Article 7(11))
- Search Engine Roundtable – Google Helpful Content Doc With New Main Content & EOT/SA Sections (2 October 2026)
- Relevant Audience – Google adds fake-author warning to helpful content page (3 October 2026)
- Search Engine Land – Google quality raters now assess whether content is AI-generated (January 2025)
- PPC Land – Google updates quality rater guidelines with AI content evaluation criteria
- Hobo Web – QualityNsrPQData and the contentEffort attribute in the Google documentation leak
Want to know how your content does in this assessment?
In an SEO audit I review the most important pages for effort, originality and accuracy, and point out the content to improve and the content that is better merged or removed.
See the SEO auditHow this text was written
I will not pretend it was written entirely by hand, by candlelight. For the research, the proofreading and the code of this page I used a language model. It is a good tool for digging through documentation and catching the typos you no longer see on the third read of your own text. The English version was also prepared with a model, from my Polish original.
What the model did not do: it did not choose the topic, form the arguments or decide what this data means for your site. That stays with me, as does the last word on what goes into the text and what comes out of it.
Where the model helped
- digging through sources and organising data
- proofreading and the English version of the text
- code, page layout and infographics
What stays with me
- the choice of topic and what is worth saying about it
- the arguments, the reading of the data and the conclusions
- fact-checking and the decision on what goes into the text