How to improve AI visibility in 30 days

Improving AI visibility means getting your brand named more often, more accurately and in a better position when people ask AI assistants about your category. It is measured on a fixed list of questions, not on one lucky screenshot.

Most teams start the wrong way round. They rewrite a few blog posts, ask ChatGPT a question once, and hope. Answers vary from run to run and from engine to engine, so a single check tells you very little. The plan below fixes the order: measure, unblock, answer, earn, then measure again.

It takes about 30 days of focused work to set up. The results take longer to settle, because engines refresh their search indexes on their own schedules and model knowledge changes only with new releases. What you get after 30 days is a baseline, a cleaner site and a list of sources you are working on. That is the foundation every later gain sits on.

Build a prompt set of 30 to 50 questions your buyers really ask, run each one in every engine you care about, and record the results. This is your baseline.

Good prompts come from real conversations, not keyword tools alone. Pull them from:

• questions prospects ask your sales team in first calls;
• support tickets that start with "can it…" or "does it work with…";
• review sites and community threads in your category;
• comparison searches, such as "best X for Y" and "X vs Y".

Group the prompts by buying stage. "What is the best payroll software for a 40-person company?" is a shortlist question. "Is [your brand] good for remote teams?" is a validation question. You will act differently on each.

For every prompt and engine, record four things: whether you are mentioned, your position in the answer, which sources the answer cites, and how you are described. Those four map to the metrics in our AI visibility metrics guide. Run each prompt more than once, because answers vary between runs. A spreadsheet works for a first pass; prompt tracking does it daily without the copy-pasting.

By Friday you should be able to say, per engine: "We appear in 12 of 40 prompts. Competitor A appears in 29. These 8 domains are cited most often." That sentence is your starting line.

Usually one of three things: a robots.txt rule, a CDN bot setting or content that only appears after JavaScript runs. Fix these before you write a single new page.

Check your robots.txt. AI companies run separate bots for search and for training. Blocking OpenAI's training bot does not remove you from ChatGPT search, but blocking its search bot does (OpenAI). Our AI crawler list has every user agent and a template.

Check your CDN. Many bot-protection settings challenge AI crawlers before robots.txt is ever read. Look at the bot logs for 403 responses to known AI user agents.

Check what is in the HTML. Fetch your pricing page, your main product pages and your FAQ with JavaScript turned off. If the prices, plan names or answers are missing, some AI crawlers will miss them too.

Check Google's basics. Google says there are no extra technical requirements for AI Overviews or AI Mode beyond being indexed and eligible for a snippet (Google Search Central). If a page is not indexed, it cannot be used.

These fixes are dull and fast. A developer can usually clear them in a day or two, and they unblock everything that follows.

Write or fix the pages that answer your lost prompts directly, starting with the prompts where competitors appear and you do not.

Look at your baseline. For each lost prompt, ask what page would answer it in the first two sentences. Often the page exists but buries the answer. A pricing page that says "contact us" cannot be quoted on price. A product page that never says who it is for cannot be recommended for anyone.

Practical rules that make pages easier to quote:

1. Answer first. Put the direct answer in the first sentence under each heading.
2. Be specific. Name the use case, the company size, the integrations and the price range. Vague claims give an engine nothing to repeat.
3. Add evidence. Numbers with sources, named standards, real limits. The original GEO study found that adding citations, quotations and statistics raised a source's visibility in generative answers by up to about 40% in its benchmark (Aggarwal et al.).
4. Write comparison pages honestly. If buyers ask "X vs Y", an honest page that says when the other option is better is more likely to be cited than a one-sided one.
5. Keep facts consistent. Your name, category, prices and founding details should match across your site, your profiles and your docs.

Google's own advice for AI features is the same as for search: unique, useful content written for people, on pages that are easy to use (Google Search Central). There is no secret format.

Find the domains engines cite for your prompts and work on getting accurately listed on them. For many categories, these sources shape answers more than your own site does.

Your week-1 baseline already has the list. Sort the cited domains by how often they appear in answers where you are missing. You will usually see a mix of:

• review and comparison sites;
• "best of" listicles from trade publications;
• community threads and Q&A sites;
• partner and integration directories;
• Wikipedia and data sources such as Wikidata.

Then work through them by type. Claim and complete your review profiles. Pitch an update to listicles that are maintained, with a short factual note on what you do and who you serve. Answer real questions in communities as a named employee, without astroturfing. Make sure your integration partners list you.

This is the slowest part of the plan and the one with the longest effect. A source that mentions you keeps working every time an engine retrieves it.

Re-run the same prompt set at day 30 and compare it with your baseline, prompt by prompt and engine by engine. Change one variable at a time where you can, so you know which fix moved which answer.

Expect uneven results. Engines that search the web live, such as Perplexity and ChatGPT with search, often reflect page and source changes sooner. Answers drawn from model memory move only when a new model is trained. Track them separately so a slow engine doesn't hide progress in a fast one.

Report three numbers to leadership: mention rate across the prompt set, share of voice against your top three competitors, and the number of cited sources that now include you. Keep the full answers on file. When someone asks "why did we drop?", the stored answer and its sources are the explanation.

If a number doesn't move, don't conclude the work failed. Check whether the engine has recrawled the changed pages yet, and whether the prompt is answered from memory rather than search.

Keep the loop running weekly: track, find the biggest gap, fix one thing, check again. Add new prompts when you launch products or enter markets, and retire prompts nobody asks anymore.

If you want the measuring done for you, Ossvy runs your prompts daily in six engines and ranks the next fixes in actionable insights. To see where you stand today, start with the free AI visibility check.

Week 1: measure 30 to 50 buyer prompts in every engine. Week 2: clear robots.txt, CDN and rendering blocks. Week 3: fix the pages that should answer your lost prompts, answer first and with evidence. Week 4: get listed on the sources engines cite. Then re-measure and repeat. For the strategy behind the plan, read our GEO guide.