In thirty days you can make your website readable for ChatGPT, Gemini, Claude and Perplexity, even without a content team. Week 1 covers access and a baseline measurement. Week 2 covers structure and schema.org. Week 3 covers three pages that can stand on their own as a citation. Week 4 covers mentions outside your own domain and the first re-measurement. Four to six hours a week is enough, as long as you decide up front what you are going to leave undone.
This plan is written for marketers at IT and B2B companies who run the website, the content and the campaigns on their own.
An AI assistant does not return a list of results. It picks a handful of sources, reads them and writes one answer. There is no second page where you still get found. You are in that answer or you do not exist for that buyer.
That changes what you have to fix. Classic search optimisation aims at where a page sits in a list. AI visibility aims at whether a model can retrieve your text, understand it and feel safe naming you as the source. That is a different set of changes, with its own measuring stick.
Two technical facts define most of your playing field. AI crawlers often do not execute JavaScript, so content that only loads client side does not exist for them. And an assistant prefers to quote a block that reads on its own, with a heading above it and the answer in the first two sentences.
One expectation to set. Count on three to six months before the effect is visible and stable. Thirty days gets you the foundation, not the result.
Start at the entrance. If your robots.txt blocks the AI crawlers, everything else is wasted work. The crawlers you can deliberately allow are GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot, Google-Extended and CCBot. While you are there, check that no noindex is left on pages you do want quoted, and that your sitemap is correct.
Then publish an llms.txt in the root of your domain. It is a short markdown file that describes in plain language what your company does and which pages matter most. It costs half a day and it is the cheapest step in this whole plan.
Finish the week with your baseline. Write down ten to fifteen questions the way a buyer asks them, not the way you name your service. Ask them in ChatGPT, Gemini, Claude and Perplexity. Per question, note which companies get named and which source URLs sit underneath. That list of sources is worth more than whether you are named yourself, because it shows you which pages decide the answer in your market.
Week 1 output: robots.txt updated, llms.txt live, baseline in a spreadsheet with questions, named companies and source URLs.
Now the content itself. Walk through your ten most important pages and check whether the text sits in the HTML instead of appearing only after a script runs. Look at the source, or switch JavaScript off in your browser. What you cannot see there, a crawler cannot see either.
Then break up the long slabs of text. Every section gets a heading that answers a question, and the first two sentences under that heading already give the answer. The supporting detail comes after. This shape is the reason one page gets quoted word for word and another does not.
Add schema.org markup where it adds something. The types that matter are Organization, Service, FAQPage, Article or BlogPosting, BreadcrumbList and Person. Review and AggregateRating only when there are real reviews behind them. Make sure your company name, address and phone number are identical everywhere, on your site and on every platform that lists you, and connect your profiles to your Organization block with sameAs. A model that is unsure whether two mentions describe the same company will name a company where that doubt does not exist.
Week 2 output: ten pages readable server side, headings rewritten as questions, Organization and FAQPage live, company details identical everywhere.
Three pages are enough to start. Pick them based on the questions from your baseline where you do not show up at all.
The first is a definition page. Explain in plain language the problem you solve, using the terms your market uses. The second is a comparison. Put two approaches or two types of solution side by side and say honestly when each one fits. Assistants like quoting comparisons, because they can take over the structure directly. The third is a process page or an extended FAQ. What happens when someone works with you, in what order, and what the client has to do themselves.
Write all three in the same shape. Question as heading, answer in two sentences, explanation after that. Keep sections short. Do not invent a number to make a point land, because a claim nobody can verify costs you exactly the credibility you are trying to build.
Week 3 output: three pages live, each with a three question FAQ block and links between them.
Your own site is one voice. Assistants also weigh what others say about you. Which platform counts depends on the type of IT company. For B2B software and SaaS it is G2, Capterra and your Google Business Profile. For IT services and managed service providers it is Clutch, Sortlist and Google. For hardware and technical components, Google works together with industry specific directories. For staffing and recruitment, Google is the anchor. Trustpilot does not work as an anchor for B2B, because consumer reviews carry different weight there.
Ask your own satisfied clients for a review, right after a moment where something went well. Never pay for one. Two to four reviews a month is a pace you can keep up for years and one that looks real, because it is.
In the same week, repeat your baseline with exactly the same questions. Four weeks is too short to conclude anything about results, but you will see whether you are starting to appear at all and which sources have shifted.
Week 4 output: two review platforms set up, three clients asked, re-measurement placed next to the baseline.
Measure two things. Citations per assistant, meaning how often you get named across your fixed list of questions, with the source URL alongside. And share of voice, meaning your share of the companies named on those questions. Those are the only numbers you can repeat honestly every month.
Keep the question list fixed. The moment you change the questions, you are comparing two different measurements and you know nothing. Measure monthly rather than daily, because the noise between two individual answers is large.
Be honest about cause. Models change. A rise can come from your work or from a model update, and the same goes for a drop. Guaranteed positions in AI answers do not exist, at any agency. A spot in an AI answer is not for sale.
Useful to know before you compare quotes. Research by De GEO Experts shows that 73 percent of Dutch GEO agencies charge between 2,000 and 3,000 euro a month. Every agency in that study offers technical audits and content creation, while 27 percent also carry out the technical optimisation themselves. At the rest, implementation, publication and authority building stay with the client. Source: degeoexperts.nl.
With that kind of advisory engagement, count on ten to twenty hours of internal capacity a month. That is the part that does not appear in the quote. We work hands on: we put the technical changes live on the client site ourselves, write and publish the content, send press releases to national media for mentions outside the client's own domain, call clients for reviews and draft the review text so the client only has to approve it, and deliver the monthly measurement per assistant with the source URLs underneath.
Count on three to six months before the effect is visible and stable. The technical steps in weeks 1 and 2 can show up sooner, because they decide whether you get read at all. Mentions outside your own domain have the longest lead time.
That is a decision per crawler. Some bots fetch content to support answers, others collect training data. You can allow or refuse them individually in robots.txt. Block everything and you cannot be cited either.
The content, the headings, the reviews and the measurement you do yourself. For robots.txt, llms.txt, schema.org and server side rendering you need someone with access to the site. In most content management systems that is half a day of work, not a project.
If you want to know where you stand before week 1 starts, run the free GEO scan. If you want your own team to learn how to do this, look at the GEO training for marketers. And if you would rather have us carry it out, read how we work for IT companies.