Start from what the check found
How to improve AI visibility depends on what your check showed, because the three common situations need three different kinds of work. Generic advice, such as write more content or get more links, wastes effort when it does not match the finding.
This page assumes you have run a check and recorded the results. If you have not, start with the check walkthrough, and run each question more than once, since answers vary. If you have results but are unsure what the four outcomes mean, read AI visibility results explained first. This page picks up where those leave off.
One framing point. Nobody controls what an assistant says, and nobody can promise a placement. What you can do is give assistants better material to work from, in the places they look, and then check whether the answers move. The work below is that, in order of how much it usually pays.
The three situations
Look at your results by question group and find which situation describes each group. You may be in different ones for different groups.
Situation one: missing nearly everywhere. For this group of questions, you are absent from most answers across assistants. The assistants are not drawing on you at all.
Situation two: present but not recommended. You are named in many answers, often in lists, but you are rarely first or put forward. You are known and not chosen.
Situation three: recommended but not cited. The assistants speak well of you, but your own pages are not the source. Reviews, directories or other people's pages are carrying you, and your site is not linked.
The next three sections take each in turn. If you are unsure which applies, count outcomes for the group: mostly absent is one, mostly mentioned is two, and recommended with few links is three. Most businesses will find situation one for their broad questions and situation two or three for questions where they are already known.
If you are missing nearly everywhere
This is the most common starting point, and the one with the largest gains available. The assistants have too little to go on, or what they have is not in the places they look.
Step one: fix the basics. Check that your own site states plainly who you are, where you operate, what you do and who you serve. Make sure the same facts appear on your listings and profiles. Remove or correct anything old. The page on why a brand is not in ChatGPT goes through the usual causes in detail, and technical blocks, such as a site that keeps crawlers out, are worth ruling out first.
Step two: write for the questions you are missing. Take the group where absence is highest and has the most value. For each question, write a page that answers it plainly in the first paragraph, then goes into the detail a buyer needs. Use the buyer's words in headings. Include specifics: who it suits, what it costs in general terms, what to watch out for, real examples. Pages that answer one question well give an assistant a sentence it can use.
Step three: look at who is cited. Open the sources the answers link to, and see what kinds of site they are. If the answers lean on local guides or review sites, your next job is to be on those, honestly.
Expect this to take weeks and months, not days. Start with the largest group of missing questions, not all of them at once.
If you are present but not recommended
You are in the conversation, and something is losing the comparison. This is a more precise problem, and often a more solvable one, because you can see who is winning.
Step one: read the winners. For the questions where someone else is recommended, find the sentences that recommend them. What reasons are given? Fit for a type of buyer, a specialty, a price band, an area, a record of results? Those reasons are the criteria the assistant is using.
Step two: check whether you give assistants those reasons. Do your pages say, in plain words, who you are best for and why? Many businesses describe what they sell and never say who it suits, so the assistant has nothing to say about fit. Add the missing statement, with evidence you can stand behind.
Step three: build the comparison material. Buyers and assistants both look for pages that set options side by side. If you have no page comparing your approach with the alternatives, write one, honestly, without attacking anyone. A fair comparison is often cited.
Step four: strengthen outside proof. Reviews that mention specifics, and mentions in independent places, are what make a recommendation safe for an assistant to give. Our page on reviews and AI recommendations covers how customer wording flows into answers. Ask satisfied customers to say what they liked in their own words, and never write or buy reviews.
If you are recommended but not cited
This is a good problem with a real cost. Assistants speak well of you, but the credit and the click go to someone else's page, and your standing depends on pages you do not control.
Step one: find out whose pages carry you. Look at the links beside answers that recommend you. Which sites are they? A review platform, a local guide, a news piece, a forum thread? List them.
Step two: make your own page the better source. For each question where you are recommended, ask whether your own site has a page that answers it as plainly and completely as the cited one does. If not, write it. If it exists, tighten it: put the answer in the first paragraph, add the specifics, and make sure it is easy to read and fast to load.
Step three: make the facts consistent. If your page and the third-party page disagree, the assistant may prefer the one it finds clearer. Align them, and correct the other where you can.
Step four: protect what you rely on. If a single outside page is doing most of the work, note it. Thank the people behind it, keep your information there current, and do not let that be your only support. Our guide on how to get cited by ChatGPT goes deeper on this side of the work.
A sensible order of work
Whatever your situation, the same sequence tends to pay best, because it starts with the cheapest and most certain gains.
Week one: fix wrong and old information. It costs little and removes things that actively mislead. Check your site, listings, profiles and any directories.
Weeks two to four: write for the biggest gap. Pick the group of questions with the most value and the most absence. Publish a set of plain, specific pages for it. Resist the urge to spread thin.
Month two: build the supporting evidence. Reviews with specifics, honest mentions in independent places, and consistent listings. This is slower and not fully in your hands, so start it early.
Month three: re-run the identical check. Same questions, same assistants, same setup, same number of runs. Compare with your baseline, using the method on our page on how to measure AI visibility.
If you changed several things at once, you will not know which one helped. That is acceptable for a small business that only needs the number to rise. If you need to learn, change one group of pages at a time, and keep the questions for the others as a comparison.
What a page that earns a citation tends to look like
Since so much of the work is writing pages, here is a plain description of what tends to work, drawn from reading the sources assistants link to, not from any guaranteed formula.
The page is about one question. The title and the main heading use the buyer's wording. The first paragraph answers the question outright, in a sentence or two, before any background. The rest of the page gives the detail: who it suits, the options, the trade-offs, real examples, and what to check before deciding. Headings are phrased as the questions a reader would ask next. A short list or table appears where the content is truly a list or a comparison, not as decoration.
The page says who wrote it and when it was updated, and the facts on it are ones you can stand behind. It has a few honest limits stated, such as when the advice does not apply. It links to other pages on your site that go deeper, and it is easy to read on a phone.
What it does not do is repeat a keyword for the sake of it, pad to a length, or avoid saying anything specific. Assistants need a clear sentence to lift. A page that offers one, and offers evidence behind it, is more useful to the assistant and to the person who clicks.
Checking whether it worked
Improvement is only real if you can show it, and the way to show it is to repeat the check you started with.
Run the same questions, in the same assistants, with the same number of runs and the same setup. Sort the answers into the same four outcomes. Compare each question group with the baseline, and look at the shares, not single answers. The counting rules are the same ones you used the first time.
Then read the sentences. A shift from a bare mention to a described, specific mention is progress even if the counts are flat. A new phrase that echoes something you published is a good sign that a page has been picked up.
If nothing moved, do not conclude the work was wasted. Slow sources take time, and some changes need a longer window. Check whether the pages are live and readable, whether the facts match across the web, and whether you aimed at the group of questions that carries the most value. Adjust one thing, and run the check again in a month.
What not to do
Some moves are tempting and tend to backfire.
Do not stuff pages with the questions. Repeating a question twenty times does not help. A clear answer to it does.
Do not buy or write fake reviews. Assistants and platforms look at the same signals buyers do, and fake ones destroy trust.
Do not mass-publish thin pages. Width without depth tends to fail in this area. A smaller number of genuinely useful pages beats a flood.
Do not chase a single screenshot. Getting named once in a test proves nothing, as our page on why AI answers change explains.
Do not drop ordinary search work. Search is changing, not ending. Pages that are clear, useful and well linked serve both Google and assistants.
Do not expect a promise. Nobody can guarantee a place in an answer. Be wary of anyone who does.
When to do it yourself and when to get help
A small business can do the first two stages alone: fixing facts and writing plain, specific pages for the biggest gap. It is mostly editorial judgment and knowledge of your own customers, which you have.
Help is worth considering when the volume is the obstacle. If your check shows hundreds of questions where you are absent, writing and publishing enough good pages by hand is a heavy load. It is also worth considering when the evidence-building stage needs sustained, patient effort that you cannot spare.
If you do hire, ask what will be published in the first month, who writes the questions, and how they will repeat the same check to show change. Our guide to AI visibility optimization sets out what moves visibility and what does not, which is a useful yardstick against any proposal.
How RunOctopus does this work
The work above, writing plain specific pages for the questions a business is missing, is what RunOctopus does for its customers. It is a done-for-you service. Ollie, its AI, researches the business and then writes and publishes pages directly on the business's own website, and a team plans, writes, reviews and publishes alongside it. It works with Shopify, WordPress, WooCommerce and Wix. There is nothing to install and nothing to approve, upload or schedule, and pages go live on their own.
In the first week a customer gets eight long-form articles, six pages and a custom interactive tool built for the niche. After that it is 175 or more pieces of content a month, written, published and linked. Reports show where ChatGPT, Perplexity, Claude and Gemini name the business, and the 90-day promise is more visibility on Google and in AI answers than in the 90 days before joining, or every dollar back. There is no free AI visibility checker.
To see the plan Ollie would make for your site, use the free site read at app.runoctopus.com/start. It reads your address and shows the plan, and it does not test AI answers. No account, no call.