How to Write a Blog Post in the Era of AI – The anatomy of a blog post



What changed: your post now competes with a summary of itself

AI answers now sit between your post and the reader, and they roughly halve clicks. When a Google AI summary appeared, link clicks fell to 8% of searches versus 15% without one, and only about 1% of clicks happened inside the summary box (Univ. of Toronto, 2025, reviewing Pew data). So being cited is not the same as being clicked.

SignalFigureSource
UK keyword searches showing an AI overview~30%Ofcom Online Nation 2025
UK online adults who read AI summaries at least sometimes75%Ofcom Adults’ Media Use 2026
UK adults using AI tools54% (31% a year earlier)Ofcom Adults’ Media Use 2026
AI-summary searches ending with no click~26%Univ. of Toronto, 2025
Respondents who click through to sources from AI chatbot answers4%Reuters Institute DNR 2026
Informational and commercial keywords showing an AI Overview98% of 50 tested, median 11 sources citedOn-Page.ai, 2026
Marketers using AI to create content87% of 879Ahrefs, 2025

But SEO isn’t dead. Google still handles around 3 billion UK searches a month and is used by 82% of online adults (Ofcom, 2025). The on-page basics still earn the ranking; what changed is what a post must do once it ranks.

Three conclusions shape everything below:

  • AI use isn’t the problem; low effort is. Google’s rater guidelines give the Lowest rating to pages whose main content is almost all AI-generated “with little effort, originality or added value” (section 4.6.6, Google QRG 2025). Only 4% of marketers publish mostly unedited AI content (Ahrefs, 2025).
  • Evidence gets cited, keywords don’t. Adding statistics, quotations and cited sources raised visibility in AI answers by 30% to 40%; keyword stuffing did about 10% worse than doing nothing on Perplexity (Princeton et al., 2024).
  • AI engines trust other people’s words about you. ChatGPT drew 93.5% of its citations from earned media for well-known brand queries (Univ. of Toronto, 2025). Your post is one input; mentions elsewhere are the other.

One honest caveat: originality is not a guaranteed AI-citation lever. AI Overviews cited the most original page on the SERP in only 33% of cases, below the 50% expected by chance (On-Page.ai, 2026, 50 keywords, one day, U.S.). Originality earns the click, the trust and the conversion; structure and evidence earn the citation. You need both.

The process: AI does the scaffolding, you supply the evidence

Use AI where the data says it helps: brainstorming topics (76% of AI users), outlines (73%) and improving existing content (67%) (Ahrefs, 2025). Keep the parts an LLM can’t fake for yourself: the experience, the data and the opinion.

  1. Start from the query fan-out, not one keyword. List the head term plus the sub-questions an AI engine will split it into (cost, comparisons, “is it worth it”, alternatives, local options). 75% of AI Overview citations pointed to pages not on page one for the head query (On-Page.ai, 2026), so covering the sub-questions well is a way in even if you don’t rank for the head term.
  2. Pick the intent, and the conversion that matches it. Commercial queries (“best”, “vs”, “cost”, “near me”, “for [use case]”) need a decision at the end of the page, not a newsletter box.
  3. Run discovery before drafting. Interview the expert (or yourself): a client problem, a decision that went against standard advice, a number from a project, a mistake. Ask follow-ups until you have specifics. No firsthand evidence yet? Don’t write the post yet.
  4. Find the non-commodity angle. Apply the interchangeability test: if five competitors could publish the same post by swapping the company name, it’s commodity content. Lead with the case study, test, dataset or contrarian opinion that only you can back up.
  5. Let AI draft the outline, then rebuild it around your evidence. Each H2 should answer one sub-question and carry at least one stat, quote or cited source. That’s the combination the Princeton GEO study found lifted visibility 30% to 40%.
  6. Write the answer first in every section. One or two self-contained sentences that still make sense if lifted out of the page, then the reasoning, then the example.
  7. Edit like a sceptic. 80% of AI users fact-check manually, 35% bring in a subject-matter expert and 30% use an editor (Ahrefs, 2025). Do all three on money pages. Strip filler intros, generic conclusions and anything that sounds like every other AI draft.
  8. Put a human on it. 57% of UK adults aware of AI would trust an AI-generated news story less than a human-written one (Ofcom, 2026). Named author, credentials, photo, date, and a clear point of view.
  9. Publish, then earn mentions elsewhere. Pitch the original data point to publications, podcasts and partners. AI engines cite earned media far more than brand sites: 64.6% earned vs 34.1% brand across bank-ranking queries (Univ. of Toronto, 2025).
  10. Measure AI visibility carefully. ChatGPT searched the web in only 21% of runs, and 85% of cited URLs appeared in just one of three identical runs (On-Page.ai, 2026). In the UK, the CMA now requires Google to provide referral data showing visits from its generative AI features (CMA, 2026).

Stop doing:

  • keyword stuffing
  • publishing unedited drafts
  • 3,000-word “ultimate guides” that restate page one
  • and chasing volume for its own sake

AI users publish more (median 17 vs 12 articles a month) on almost the same budget (about $2,475 vs $2,442 a month) (Ahrefs, 2025). Speed is now cheap, so it’s no longer a differentiator.

Anatomy of a post that converts and gets cited by AI

A commercial post has two readers: the AI engine that extracts a passage, and the buyer who clicks through to decide. Every block below has to serve one or both. Top of page to bottom:

13 elements for Google Search, AI Overviews and AI assistants, tagged by who each one serves

Buyer: drives the conversionAI: earns the citationBoth
1
1

Title tag, URL and H1

Both
  • Commercial query plus its modifier (“vs”, “cost”, “best”)
  • Promise a decision, not a topic
  • Short, readable, hyphenated URL
2
2

Byline and author box

Both
  • Real name, role and why they know this
  • Last-updated date, reviewer if any
  • Link to a full author page

57% of UK adults aware of AI trust an AI-written story less than a human one (Ofcom, 2026)

4
4

The proof hook

Buyer
  • Your case study, test or dataset
  • Dates, sample size and outcome
  • The thing five competitors can’t copy
7
7

Comparison table

Both
  • Options × the criteria buyers weigh
  • Mark your pick and say why
  • Live HTML, not an image
8
8

Mid-page CTA

Buyer
  • Tied to the section just above it
  • Catches readers who decided early
  • One action, no pop-ups
9
9

Original video or visual

Both
  • Your own demo, crew or chart
  • Not stock photography
  • Caption it with what it proves

YouTube took 22% of AI Overview citation links in one sample (On-Page.ai, 2026)

10
10

Limitations

Buyer
  • Who this is not for
  • Honest trade-offs and exceptions
  • Filters out bad-fit leads
11
11

FAQ

Both
  • Real questions from sales calls and support
  • 2–3 sentence answers
  • Maps to long-tail prompts
12
12

Final CTA and proof

Buyer
  • One clear next step
  • A real review or result beside it
  • Match the offer to the intent

1 · Rank

On-page basics still matter: Google handles ~3bn UK searches a month (Ofcom, 2025)

2 · Get cited

Answer first, evidence in every section, mentions on other sites

3 · Convert

Give the click a reason: data, tools and a next step the AI summary can’t

Sources: Aggarwal et al. (Princeton) 2024 · Chen et al. (Univ. of Toronto) 2025 · On-Page.ai 2026 · Ofcom 2025–26.Luca Tagliaferro · lucatagliaferro.com

Read it top to bottom: the proof sits in the first screen, and each block is tagged with the reader it mainly serves. The table below gives the detail for each block.

#BlockJob for the buyer (conversion)Job for AI engines (citation)What to put in it
1Title + H1Promise a decision, not a topicMatches the commercial query and its modifier“Best X for Y in 2026: what we found after testing Z”; the head term plus the qualifier
2Byline + author boxTrust: a real expert made thisE-E-A-T signal; entity link to the authorName, credentials, photo, last-updated date, how you know this
3Answer-first summary (40–80 words)The verdict in 10 secondsThe passage most likely to be liftedThe direct answer, one number, who it suits and who it doesn’t
4The proof hookWhy believe you over page oneUnique, quotable informationYour case study, test or dataset, with dates, sample and outcome
5Fan-out H2sAnswers each objection in orderEach H2 matches a sub-query the engine runsQuestion-style H2s: cost, comparison, alternatives, risks, “is it worth it”
6Evidence in every sectionRemoves doubtStats, quotes and citations lifted visibility 30–40%One stat, one named-expert quote or one cited source per H2
7Comparison tableSpeeds the decisionStructured data is easy to extractOptions × criteria that matter (price, fit, trade-off), your pick marked
8Mid-page CTACatches readers who decided early—Contextual, tied to the section above (“Get a quote for this setup”)
9Video or original visualShows the workYouTube took 22% of AI Overview citations in one sampleYour own demo, walkthrough or chart, not stock
10Limitations and who it’s not forCredibility; filters bad leadsBalanced passages read as trustworthyHonest trade-offs, exceptions, when you’d advise against it
11FAQClears the last objectionsShort Q&A pairs map to long-tail promptsReal questions from sales calls and support, answered in 2–3 sentences
12Final CTA + proofConverts—One clear next step, plus a testimonial or result next to it
13Schema, internal links, sources—Machine-readable context; trust in claimsArticle/Product/FAQ schema as relevant, links to money pages, a sources list

Evidence for the rows: answer-first and evidence blocks from Princeton et al., 2024; byline and trust from Ofcom, 2026; video from On-Page.ai, 2026. The rest is conversion practice, not research from the dossier.

What changes for commercial keywords

  • Earn the click after the answer. With only ~1% of clicks inside the AI box, the summary will give away the generic answer. Your page must hold what the summary can’t: the full comparison, the calculator, the pricing breakdown, the case study, the quote form.
  • Lead with first-hand testing. A year-long study of 7,392 queries found most high-ranking product reviews used affiliate marketing (Leipzig / Weimar, ECIR 2024), which is part of why Google now pushes first-hand reviews. Show photos, numbers and what went wrong.
  • Small brands need third-party mentions. Perplexity named major brands in 67.9% of brand mentions for unbranded cola queries versus 5.8% for niche brands (Univ. of Toronto, 2025). Your post should contain the stat or line others will want to quote, and then you go and get it quoted.
  • Lower-ranked pages gain most. Citing sources raised visibility 115.1% for pages ranked fifth, while the top page fell 30.3% (Princeton et al., 2024). Structure and evidence can level the field.
  • Write plainly. Readers needed about 13 years of schooling to understand half the web content studied (Univ. of Porto, dated data). Short sentences convert better and extract cleaner.
  • Quality is a conversion issue, not just a ranking one. 65% of marketers rate human-written content as better, yet most think both rank the same (Ahrefs, 2025). Rankings may forgive a bland post; buyers won’t.