Training template · open
Writing for AI search
How to plan, write and upload content that AI Overviews, ChatGPT, Perplexity and Gemini can read, trust and quote. The open template of the 60-minute session I run for content teams and project managers: 27 slides.
This is a template, not the training
Every training I run is tailored to the client: your products, your category, your markets and the people in the room. We work on your pages, your questions and your competitors, not on these generic examples.
Use this deck to see the approach. The live session, with your examples, exercises and questions, is a different thing.
Writing for
AI search
How to plan, write and upload content that AI Overviews, ChatGPT, Perplexity and Gemini can read, trust and quote. A hands-on session for project managers and content teams.
Short talk, long conversation
01 · What changed
How people search now, and why "ranking" no longer equals "being seen".
02 · How to build content
The writing rules, page anatomy and content blocks that get pages quoted.
03 · Process & measurement
Who does what, the upload checklist, and how to track it (tool walkthrough).
Your questions
Two quick discussion stops along the way, then open Q&A on your own projects.
People don't search any more. They ask.
- 2–3 keywords, no context
- Ten blue links, we compare ourselves
- Several clicks, several tabs
- Visibility = ranking position
- Long, situational questions — and follow-ups
- The answer and the shortlist arrive before any click
This is mainstream behaviour, not early adopters
Ranking is not the same as being cited
keywords with an AI Overview where the brand's site was actually cited — about 6%. It ranked in the top 10 for far more.
- Impressions flat, clicks roughly halved
- Content good enough to help the AI answer…
- …but not structured well enough to get named
For very dry, cracked hands, dermatologists usually suggest a fragrance-free cream with urea or glycerin, applied after every wash and at night.
Look for: 5–10% urea for rough skin, ceramides for the skin barrier, and a richer texture for night use.
How an AI answer gets built
One question in
"Best cream for a baby's nappy rash?" — often long and personal.
Split into sub-questions
The AI quietly asks 5–20 smaller ones: causes, how often to apply, when to see a doctor, which ingredient, price…
Grab passages, not pages
For each sub-question it pulls the smallest chunk that answers it — often via Google or Bing's index.
Write & cite
It stitches the answer together and credits sources that were clearest and most trustworthy.
Your turn
What actually changes in how we write
| Before | Now | |
|---|---|---|
| Starting point | One keyword per page | A map of questions people ask around the topic |
| Unit of content | The whole article | Self-contained answer units — each section quotable on its own |
| Opening | Warm-up intro, brand story | The answer in the first 2–3 sentences |
| Headings | "Pricing", "Benefits" | The real question: "How much does it cost?" |
| Proof | Claims, adjectives, banners | Facts, numbers, named sources, a named expert — as text |
| Success | Rankings and clicks | Citations, brand mentions, share of voice — and clicks |
Build the question map first
- Ask the AI: "What questions does someone searching for [topic] ask? List 15."
- Add real data: People Also Ask, Search Console queries, customer service, reviews, Amazon Q&A.
- Tick off what the page already answers clearly and early — not in paragraph five.
- The gaps are your brief: new sections, headings, FAQ questions.
At scale: fan-out simulators (e.g. Qforia), keyword + prompt research tools.
| Type | Example |
|---|---|
| Problem | "Why does my skin crack on my hands in winter?" |
| Situation | "I wash my hands 20 times a day at work — what helps?" |
| Recommendation | "Which hand cream works best for very dry skin?" |
| Comparison | "Urea cream or panthenol — what's the difference?" |
Rule for PMs: map every important question to one page. No page for it = a content gap. Two pages for it = cannibalisation.
Eight rules that make a passage quotable
Answer first
Open every page and section with the conclusion. Context comes after.
Headings as questions
Phrase H2/H3 the way people ask — the model matches them to the prompt.
One idea per section
Each block makes sense if lifted alone. No "as mentioned above".
Facts, not slogans
Who it's for, who it's not for, how much, how long, how often.
Text, not images
Claims inside banners and infographics are invisible to AI. Write them out.
Plain language
Short sentences. Name things explicitly; define terms on first use.
Evidence
Numbers, named sources, certifications, first-hand experience.
Kept fresh
A visible, honest "last updated" date — and real updates behind it.
Two layers: the foundation and the citability layer
Title, navigation, headings, specs, related links, CTA, footer. Most sites have it. Necessary — but it wins nothing, because competitors have it too.
A direct answer, a definition, a named author, sourced numbers, a comparison, a real FAQ. Some are new blocks, some are foundation sections rewritten. Almost nobody has this yet — this is where the gain is.
A menu, not a checklist
Instant answers
Near the top — what AI reaches for first.
Trust & evidence
Why a model believes you over others.
Decision & depth
Where brands out-answer retailers.
Guided
Native format for "how do I…" and voice.
Pick by page type. Article: quick answer + takeaways + FAQ + sources. Product page: at a glance + who it's for + how to use + FAQ. Category: decision guide + comparison.
Design test for every block: what exact snippet would an AI lift from this — and would a busy reader get it in five seconds? If neither, it's decoration.
Same topic, same facts — rewritten to be quoted
Benefits → Is it safe for newborns?
Apply regularly → Apply a thin layer at every nappy change, on clean, dry skin.
See a specialist if needed → Weeping or bleeding patches → see a doctor; dry and flaky → moisturise twice a day.
The page where people buy is usually the least AI-ready
- Open with the problem it solves, not the brand name or heritage story.
- "At a glance" facts in text: for whom, key ingredient/spec, use for, format, free from.
- Who it's for — and who it's not for. It makes recommendations precise.
- How to use as numbered steps with real quantities and frequency.
- Comparison inside the range ("which one is right for me?").
- FAQ from real customer questions, reviews and Q&A.
- Claims in A+/banners also written as text — "Since 1772" in an image doesn't exist for AI.
Title → Bullets → A+ text → Attributes → Reviews → Q&A / FAQ
Text inside images comes last, if at all. Amazon's assistant (Rufus, now "Alexa for Shopping") answers from these fields.
Most common gap we find in audits: no FAQ at all, and A+ content that is 100% images.
Make expertise visible on the page
Named author
Name, role, short bio, photo, profile link. "Dr Anna Nowak, dermatologist" beats "Editorial team".
Expert reviewer
For health, finance, safety topics: "Reviewed by…" plus the review process.
Last updated
Visible date, kept honest. Undated content reads as outdated to people and models.
Named sources
2–3 credible, linked references. Original data and first-hand testing are the most quotable of all.
The form changes — the facts never get invented. Every claim must trace back to an approved brand source. That's what keeps regulated work (pharma, OTC, finance) safe — and fast to approve.
Common mistakes & over-engineering
- Bolting on an "AI section" while the page itself stays unchanged.
- Every block on every page — a 40-question FAQ under a product nobody asks 40 things about.
- Generic AI-written filler: it says what every other page says, so there's nothing to cite.
- Repeating the keyword instead of answering the question.
- Key copy only in images, tabs or JS widgets that crawlers never see.
- Dates changed without real updates — trust-killer when caught.
- "Improving" facts during the rewrite — and re-opening legal/medical review.
- Chasing hacks (llms.txt, hidden prompts, "AI-only" pages). There's no switch.
AI builds its answer from the whole category
| google.com | 118 |
| Medical association site | 42 |
| Online pharmacy #1 | 39 |
| Brand's own site · #4 | 26 |
| Online pharmacy #2 | 26 |
Source: Chatbeat baseline, Sept 2026
- Retailers, pharmacies and marketplaces are often the source the AI prefers — and they describe our product their way.
- Reviews, forums, Reddit, media corroborate (or contradict) what we say.
- What content teams can control: the same facts, names and claims across brand.com, syndication feeds, retailer PDPs and Amazon.
- What PMs can flag: outdated ranges, old prices, discontinued products still live on partner pages.
Technical SEO = technical GEO
- Indexed in Google and Bing — many AI answers are grounded in their results.
- AI crawlers not blocked by robots.txt, CDN or firewall (OAI-SearchBot, PerplexityBot, Bingbot…).
- Key content in the HTML, not loaded only after JavaScript or behind logins/pop-ups.
- Real heading tags (H1–H3), lists and tables — not styled divs.
- Structured data that matches the visible page: Article, Product, FAQPage, Organization, BreadcrumbList.
- Fast, mobile-friendly pages.
Very little, for now. A few new bots to allow, and log files worth checking for AI crawler activity. llms.txt exists, but no major AI search engine has confirmed using it.
If the site is technically healthy for Google, it is ~95% ready for AI search. The difference is made by content — which is why this session is about content.
Pick one page you own
Who does what
- Kickoff: priority pages, markets, approvals
- Collect author/reviewer details early
- Brief starts with the question map
- Book review rounds (legal / medical)
- Visibility baseline & prompt set
- Question map & gap analysis
- Page blueprint per template
- Schema spec, QA, reporting
- Answer-first, question headings
- Facts only from approved sources
- Fill blocks that fit the page type
- Run the pre-publish checklist
- Brand, legal, medical (LMR) check
- Confirms author / reviewer
- Fewer rounds when facts don't change
- Structure survives the CMS
- Schema, dates, author fields filled
- Final check on the live page
- Notifies for re-measurement
Where good content gets lost: the CMS
- Headings are real H2/H3 tags — not bold text, not a bigger font.
- Only one H1, matching the page's main question.
- Quick answer sits directly under the H1, above hero images and banners.
- Tables are HTML tables, never screenshots of tables.
- FAQ is visible text in the HTML (accordion is fine if content loads with the page).
- Author, reviewer and "last updated" fields filled and visible.
- Schema added and valid — and matches what's visible (Rich Results Test).
- Alt text describes what the image shows, captions on charts.
- Internal links with descriptive anchors — not "click here".
- Check the live page on mobile and "view source": is the text really there?
How we know it worked
Google Search Console
Clicks, impressions, CTR, position — before vs after, per optimised URL.
Bing Webmaster Tools
AI performance data for Copilot and Bing-grounded assistants.
GA4
Referral traffic from chatgpt.com, perplexity.ai, gemini.google.com as its own segment.
Prompt tracking
Chatbeat (by Brand24) and similar tools: mentions, position, sentiment, cited sources per engine.
50–100 prompts · no brand names in them · written as the buyer · every prompt able to name a product · each mapped to a page · frozen for at least a quarter.
Recommended (is the brand named?) and cited (is our page a source?) are different. Strong in one engine, absent in another is normal.
Reading a visibility dashboard
Brand score — how often and how high the brand appears.
Visibility — share of all answers that mention the brand.
Average position — only counts answers where the brand appears, so it flatters.
From dashboard to content brief
| Which hand cream works best for very dry skin? | 67% |
| Urea or panthenol for cracked hands? | 33% |
| Best hand cream for nurses who wash hands a lot | 0% |
| Fragrance-free hand cream for sensitive skin | 50% |
| Hand cream for eczema on fingers | 17% |
For very dry skin, Brand A Intensive is a strong pick: 10% urea and ceramides, fragrance-free. Brand B is a lighter daytime option; Brand C suits eczema-prone skin.
Zero-visibility prompts = the content backlog. In the pilot: informational prompts 0/20 in ChatGPT.
Cited sources = where else we need to be present and consistent (retailers, associations, media).
How we're described = qualitative check: right facts? right positioning? outdated info?
What moves fast, what takes time
On-page changes picked up
Re-crawl and re-indexing of optimised pages; first citation changes in Google AI Overviews and Perplexity, which read the live web.
A readable trend
Monthly prompt tracking shows direction. Single runs are noise — answers vary run to run and by user.
Durable visibility
Comes from consistency: fresh content, consistent facts everywhere, and trust built off-site over time.
The pre-publish checklist
- The page answers a mapped list of real questions, not one keyword.
- The answer is in the first 2–3 sentences and on the first phone screen.
- Headings are questions, and each section opens with its answer.
- Every section makes sense on its own.
- Key facts and claims are in text, not only in images.
- At least one depth block where it fits: comparison, steps, do's & don'ts.
- FAQ with genuine long-tail questions and self-contained answers.
- Author, reviewer, last-updated date and named sources visible.
- No fact added that isn't in approved source material.
- Live page checked: real H tags, HTML text, valid schema.
Questions?
- Which of your current projects could use a question map?
- What blocks us most: client approvals, CMS limits, or source material?
- Which pages should we baseline first?
- learningseo.io — Optimize for AI Search roadmap
- Google Search Central — Top ways to ensure your content performs well in AI experiences
- Aleyda Solis — AI Search Content Optimization Checklist
- tomek-sikora.com — query fan-out, citable pages, prompt sets, GEO without a blog factory
Who this session is for
Content teams, copywriters, editors and the project managers who brief them. It is practical: everything on the slides can be applied to the next brief, article, product page or upload, with two short discussion stops and an exercise on a page the team owns.
It is one module in my SEO & AI Search training. For the bigger picture, start with AI search: the basics.
Common questions
Questions about this training
-
It is the full structure and the generic content. The live session is tailored to your category, pages and team, uses your own content as examples, and includes discussion stops and an exercise.
-
Content teams, copywriters, editors, content uploaders and the project managers who brief them.
-
About 60 minutes: roughly 35 minutes of content and 20 or more minutes of questions. It can be extended to a half-day workshop on your own pages.
-
No. Everything on the slides can be applied to the next brief, article or product page. Measurement uses Search Console, GA4 and, optionally, an AI-visibility tracking tool.
Work together
Want this training for your content team?
Tell me what you publish and who will be in the room. I’ll build the session on your own pages and briefs.
Open for projects
