Blog content, SEO, and GEO: build the answer before somebody searches for it

You answer a good question on a customer call. A week later, somebody asks the same thing by email; then it appears in a community, followed by a third version in a direct message, and somehow your website still doesn't answer the question.

That is usually the right moment to start a blog. It's something that should come fairly early, while faster channels are still teaching you what customers care about, because a public body of work needs time to accumulate: you're building an owned library of answers to real buyer questions. Search engine optimization (SEO)The work of helping search engines discover, understand and present a website's pages to people making relevant searches. helps search systems discover and show those pages; generative engine optimization (GEO)The work of making accurate material easier for answer engines to retrieve, represent and cite, using clear information, traceable evidence and accessible pages. concerns whether answer enginesA system that responds to a question with a generated answer assembled from retrieved or learned information, sometimes with links or citations to its sources. such as ChatGPT, Claude, Gemini and Perplexity can retrieve, represent and cite them.

Choose your territory and collect the questions

A content strategy decides which subjects your content will keep returning to and why those subjects help a potential customer. Its recurring subjects are often called content pillarsA bounded subject the business returns to because buyers ask about it, the founder can cover it credibly and a better answer helps somebody understand the problem, product or choice.. Three to five of them is where you usually begin, and each should pass three tests:

1. Buyers ask questions about it before or during a purchase.
2. You have experience, evidence or access to credible sources.
3. A better answer would help the right person understand the problem, your product or a choice they need to make.

“AI” and “marketing” are topics, not pillars. A bounded pillar such as “how solo founders use AI without surrendering judgment” is narrow enough to produce related questions. “Founder thoughts” isn't a pillar unless your customers have formed a very unusual purchasing committee.

Build the question bank from what people already ask in sales conversations, support messages, interviews and communities. A repeated question from a real person beats an impressive keyword with no obvious connection to the product. You only need to record enough to understand the question and choose whether it deserves an article:

“How do I know when to reorder?” may sound like a simple how-to question, but a buyer who currently judges stock by eye may be deciding whether to trust a new system. Light keyword research can then show you how people phrase the question and what the current results miss.

Match the article to the decision

Five types cover most of the first question bank. The best choice is whichever helps the reader make the decision.

A personal story may support any of these when it changes what the reader can infer. It doesn't need its own slot in the calendar, and it certainly doesn't need to finish with five lessons about leadership (save that for LinkedIn).

Find your answer before AI finds one for you

Write a rough answer in your own words before asking AI for an outline. It captures what you currently think, and your posts need a true human voice. Then split into four groups:

That evidence ledgerA record separating verified external facts, first-party evidence, recommendations and unresolved claims before an article is drafted. stops a sourced fact and your opinion from blending into one confident mush, which is particularly easy when an AI tool has filled the gaps with dispatches from the Republic of Probably.

The best-known early GEO study, written by researchers from Princeton, IIT Delhi and elsewhere, shows why the limits belong beside the finding. The team tested content changes across 10,000 queriesThe words a person enters into a search engine or answer engine to ask for information. and found that sources, quotations and statistics improved its experimental visibility measures. The maximum improvements were 41% on one measure and 28% on another, but the study used the top five Google results, GPT-3.5 and a specific generated-answer setup; its Perplexity results differed. It supports testing evidence-rich writing, not the popular promise that adding one statistic will increase your ChatGPT citations by 41%.

Write one page for people, search and answer engines

SEO and GEO have different measurements, but they're close enough in practice that you don't need to split out into separate strategies. Google's May 2026 guidance says its generative search features use the existing Search index and ranking systems, and favors original material with first-hand experience or a distinctive point of view over another summary of what is already online.

Four writing moves make a factual article easier to understand and verify:

1. Give the answer where the question is resolved. A sentence should still make sense if somebody quotes it without the preceding three paragraphs.
2. Put attribution beside the claim. Name the source, date and population where they affect how the number should be read.
3. Define an unfamiliar term when the reader meets it. Definitions help when the language is confusing; they aren't an excuse to coin six proprietary categories.
4. Use real limits and numbers. A time, threshold, count or range helps only when you can show where it came from.

The structure should follow the information. It gives readers a faster route to the part they need and gives an answer engine a cleaner piece of material to extract.

These formats aren't citation tricks. A list helps when the material belongs in a list; it becomes irritating when every paragraph has been chopped into bullets. You don't need all four writing moves in every section or every heading rewritten as a question.

Romy's article on AI-blindness covers the familiar signs of generated prose, but removing a few phrases can't supply the missing judgment. Capture your view and examples before drafting, then read the result aloud. GEO should make the source easier to understand without making you sound like a robot.

Romy also publishes free writing and GEO skills. GEO Optimised Content can help you structure a piece of content in the preferred format for LLMs, and Personal Voice builds a profile from existing source material.

Make the page reachable, then ignore the hacks

Good writing can't help if the page is unavailable to readers and crawlers. The publication check is easier than you might think:

These checks make a page eligible to be found, and using these checks with your LLM can make this an easy task.

One article should create more than one piece of work

A blog still needs distribution, so give every article exposure through at least one channel you already use. It can then supply other channels, provided you adapt the idea instead of copying the first 300 words everywhere.

You don't owe the internet five assets for every article. Some deserve one email; a strong research piece may supply several posts over time. The adaptation only needs to give its audience a complete thought. The content spreadsheet can survive an empty cell.

Publish one strong article a week

One article a week is a reasonable starting routine - publish more only when you have the capacity and material. The weekly work is:

1. Select one recorded buyer question.
2. Confirm its content pillar, underlying decision and article type.
3. Research the missing evidence and build the evidence ledger.
4. Draft, fact-check and run the voice and GEO review.
5. Complete the technical publication check.
6. Choose the initial distribution route and any worthwhile adaptations.
7. Record the baseline, then return to the results over time.

Measure three different ways the article can work

Search discovery, AI representation, and customer action are related routes.

Google's Search ConsoleGoogle's service for checking how a website is crawled, indexed and shown across Google Search, including its generative-AI features. has reported which pages appear in its generative-AI features since August 31, 2026. OpenAI adds utm_source=chatgpt.com to ChatGPT search referral links, so those visits can be identified in web analytics.

For other answer engines, begin with five buyer-like questions and record the responses from the same models once a month. One answer from one model is an anecdote with excellent typography. When the manual check becomes tedious, Romy's guide to tracking AI visibility for free with Lettertrace explains an open-source option that runs prompts repeatedly and reports the variation.

For B2B software, G2's April 2026 study of more than 1,000 buyers found that 54% named generative AI chatbots as an influence on their shortlist, compared with 43% for review sites and 36% for vendor websites. G2 sells into this shift and the research only covers B2B software, but it suggests that the description inside an answer may matter before a visit appears.

Complete your first answer-page brief

Your output from this chapter is a first answer-page brief. Complete the record, then attach it to the following prompt with your positioning, permitted customer notes and trusted sources.

I'm preparing my first answer-page brief. Use the attached record to organize
and review the material, but don't draft or publish the article or take action
on my website, CMS, email account or social channels.

Return:
1. The reader's question, the decision underneath it and my short answer.
2. An evidence ledger separating verified facts, first-party evidence, my
   recommendations and unresolved claims.
3. The best-fitting article type, a connected outline and the search language
   supported by the material. Don't invent search volume.
4. Any clear-answer, attribution, definition or specific-number opportunities
   the article has earned, plus any steps, checklist, bullets or comparison
   table that would make the material easier to use.
5. Five stable, buyer-like prompts for checking AI visibility over time.
6. A technical publication checklist, one initial distribution route, fitting
   channel adaptations and blank fields for the three types of outcome.

Use my language where it's clear. Don't invent sources, quotations, numbers,
results, experience, demand or technical facts. Mark anything that needs
verification and leave unknowns unresolved. Stop with the brief for my review.

Preparing the brief completes the chapter; drafting and publication remain your decision.