Content & trust

Google stopped counting your keywords years ago. It started weighing your trust.

If your content plan still revolves around repeating a phrase enough times, you are optimising for a search engine that was retired while nobody was watching. Modern search reads meaning, checks who is speaking, and increasingly answers the question itself. This page explains what replaced the keyword, and why the publishing platform you choose, and the honesty you publish with, are now ranking decisions.

What Google reads now Why we publish on Ghost

Argument one

From counting words to weighing meaning.

Early search engines were librarians with a tally counter: say "plumber Boksburg" often enough and you ranked for it. That world invited an arms race of stuffed pages, and Google spent two decades dismantling it. The pivotal shift was from keywords to entities: an entity being a thing the machine recognises as real (a business, a person, a place, a product) with known relationships to other things. Google's Knowledge Graph does not see the string "AutoHive"; it sees a company, in South Africa, with divisions, services and a parent organisation. Your content now ranks for what it is about, not the words it repeats.

Then came the language models. BERT (a model Google announced in 2019 that reads words in the context of the words around them) meant search finally understood that "flights to Durban" and "flights from Durban" are different questions. MUM, announced in 2021, extended that understanding across languages and formats. And the destination of that road is visible on results pages today: AI Overviews, where the engine composes an answer from sources it trusts and cites them. The strategic consequence is blunt: you are no longer only ranking for a click; you are auditioning to be quoted. The same logic serves whoever asks the question through a chatbot instead of a search box.

Meanwhile the old trick became a confession. Keyword-stuffing is named in Google's own spam policies: density that reads unnaturally no longer looks like relevance, it looks like manipulation, and machines that read context can tell the difference. Writing badly on purpose is now a negative signal. There is a certain justice in that.

The old SEO playbook compared with what modern search rewards
The old playbook said	Modern search actually rewards	Why it changed
Repeat the target phrase; hit a keyword density.	Cover the topic so well the phrase occurs naturally, or not at all.	Language models read meaning in context; unnatural repetition now reads as spam, not relevance.
Publish a thin page for every keyword variation.	One authoritative page per question, deep enough to be the cited answer.	Entity-based search collapses variations into one intent, and AI Overviews cite one source, not forty near-duplicates.
Anonymous content is fine; volume is the strategy.	Named authors with demonstrable experience, on a site that says who is behind it.	E-E-A-T makes identity and first-hand experience part of the evaluation. Anonymity forfeits three letters of four.
Buy links anywhere; more is better.	Earn citations from sources your industry already trusts.	Authority is measured as reputation, and manufactured reputation is a pattern that is detectable, and penalised.
Content exists to capture the click.	Content exists to be the answer, sometimes quoted before any click happens.	Answer engines compose responses from trusted pages. You are auditioning to be quoted, not just listed.

The trust rubric, letter by letter

E-E-A-T: how Google's raters are told to judge you.

Google publishes Search Quality Rater Guidelines: instructions for the human evaluators whose judgements help calibrate its systems. Their central rubric is E-E-A-T. It is not a score you can buy; it is a description of what credible content looks like.

Experience

Has the author actually done the thing? First-hand accounts carry weight that summarised research cannot: the workshop you ran, the install you photographed, the client project with real numbers. This is the letter AI-generated filler can least fake, which is exactly why Google added it.

Hear the voice version of this argument

Expertise

Does the author demonstrably know the field? Depth, precision, and correct use of the discipline's own language: the difference between an article about tax written by an accountant and one written about accountants. Credentials help; demonstrated command of detail helps more.

Ask how your content reads

Authoritativeness

Do others treat you as a source? Citations, mentions, links from respected sites, a consistent identity across the web: reputation as seen from outside. It compounds slowly and cannot be bought in a month, which is why reputation management is a discipline and not a campaign.

Read the reputation discipline

Trustworthiness

The umbrella over the other three, and by Google's own account the most important: is the page accurate, honest about who is behind it, and safe to act on? Clear authorship, working contact details, cited sources and corrected errors all feed it. Trust is the ranking asset. The rest are evidence for it.

See how we stay auditable

So the working method is a sentence: write for the reader, structure for the machine. The reader gets plain language, honest claims and first-hand substance. The machine gets three specific courtesies, and this very site extends all three, on every page.

Answer blocks

A short paragraph near the top of the page that answers the page's central question directly and completely: the exact shape an AI Overview or answer engine wants to quote. Humans get their answer without scrolling; machines get a clean extract. View this page's source and you will find one.

See the home page's answer block

Schema markup

Structured data means labels in the page's code declaring what it is: an article, a business, a price list, a set of questions. Invisible to the reader, unambiguous to the crawler. It is the difference between hoping the machine infers correctly and telling it outright.

Read the machine-readable brand argument

llms.txt

A plain-text file at the root of a site that tells AI language models what the site contains and where its canonical answers live: a courtesy map for the answer engines that increasingly speak on your behalf. Young standard, trivial cost, and the sites that adopt it early get described on their own terms.

Ask us to set yours up

None of this structure earns anything if the site is too slow to crawl or land on: the same speed argument that governs your ad costs governs your visibility here, which is one reason the platform choice below is not cosmetic.

Argument two

Why we publish on Ghost, and not on the platforms everyone defaults to.

A publishing platform is not a neutral container. It decides how fast your pages load, how clean your markup is, who owns your subscriber list, and what happens if you leave. Ghost, the open-source publishing platform run by a non-profit foundation, wins on every one of those axes, for reasons that rhyme with everything else on this site.

Open source, self-hostable

The code is public and the platform can run on your own instance: your content, your subscriber list, your data, resident where you need it under POPIA. It is the same sovereignty argument as our CRM: if you ever leave us, the publication and every reader on it leave with you.

Read the sovereignty argument

Sub-second pages

Ghost ships lean pages that load in well under a second on ordinary connections. Speed is a ranking factor, a reader-patience factor, and, because landing-page experience prices your adverts, a cost-per-click factor. One decision, three dividends.

See how speed prices your ads

Newsletters & memberships, native

Email newsletters, paid subscriptions and member management are built in, not bolted on through a third-party service that owns your list. The audience you build is a first-party asset, which matters more every year as rented audiences get more expensive to reach.

Keep your audience as an asset

Markup machines can read

Ghost outputs clean, semantic HTML with structured data as standard: precisely the legibility that answer engines and AI Overviews reward. Everything argument one demands, the platform produces by default instead of by plugin.

Revisit what the machine reads

No plugin-rot

The common alternative is a core system kept alive by dozens of third-party plugins: each one a licence fee, an update risk, a security surface and a speed tax. Ghost's essentials are in the core, maintained by one team. Fewer moving parts; fewer three-a.m. surprises.

See what content on Ghost costs

Built for publishing, only

Ghost does one job. It is not also trying to be a shop, a page-builder and an app platform. Tools with one job age well, and a focused writing experience means the person with the expertise actually writes, which is where E-E-A-T's first two letters come from.

Request the content review

The conclusion both arguments point to

Trust is built the way academia builds it.

Strip away the algorithms and E-E-A-T is describing something old. Academia solved the trust problem centuries before the internet re-broke it, with three habits any publication can adopt. Citation: claims are attributed to named sources, so the reader can check rather than believe. When we reference a finding on this site, we name the researcher or the document, and when a number would be a guess, we leave the number out. An uncited statistic is not evidence; it is decoration. Primary sources: link the actual study, the actual policy document, the actual data, not a blog post summarising a tweet about a press release. Every hop away from the source adds error and shed nuance. Published corrections: when academia finds an error, the correction is public and attached to the original. We do the same: corrected posts say what changed and when. It feels uncomfortable and it is quietly magnetic, because a publication that visibly corrects itself is one whose uncorrected claims you can finally believe.

Notice that these habits are also, letter for letter, what Google's raters are instructed to look for. That is not coincidence: the guidelines are an attempt to encode, at scale, how humans have always decided whom to believe. Write like a careful scholar with a good editor, and you are optimised for the machine because you are honest with the human, not instead of it. Convincing without manipulating is not just the ethical position; on the modern web it is the winning one.

And trust compounds across media. The written word is where you demonstrate rigour; the human voice is where you demonstrate that a real person holds these views: hours of it, unedited. The two together, distributed through the right platforms and formats, are the whole content strategy. Everything else is scheduling.

One clear next step

Get an honest read on your current content.

Send us your website or blog and we will tell you, plainly, how it reads to a modern search engine: whether it answers questions or repeats phrases, whether a machine can parse it, and whether anything on it would survive a citation check. If the honest answer is "your content is fine, spend the money elsewhere", you will get that answer. This site was built by worker bees, but nobody here is a drone reading a script.

Request the content review

Or read what content on Ghost costs monthly. The numbers are public.