Ask ChatGPT for "the best invoicing tool for freelancers" and it will name four or five products. Yours is either on that list or it is not, and for a growing number of buyers that short list is the whole search. The good news is that getting onto it is less mysterious than the industry selling "AI visibility" packages would like you to think. For questions like that one, the assistants mostly search the web, read a handful of pages, and repeat what those pages agree on. So the job is to be present and described the same way on the pages they read.
That job has a name now, generative engine optimization, or GEO. It is not a replacement for SEO. It is SEO plus some attention to who else is talking about you, and this guide covers the five parts of it that actually move results, in the order worth doing them. We run a software directory, which happens to be one of the kinds of page assistants read when they build those lists, so some of this is written from the inside.
How does ChatGPT decide which tools to recommend?
An assistant can know about your product in two ways. The first is memory: whatever was in its training data, frozen at its cutoff date. The second is live retrieval: it searches, opens a few results, and writes an answer from them, usually with citations. For anything that sounds like a buying question ("best", "alternatives to", "vs", "for small teams"), ChatGPT, Perplexity, Claude, Gemini and Google's AI Overviews all lean heavily on retrieval, because product lists go stale fast.
That matters because it changes what you can influence. You cannot edit a model's memory, and it only updates when the model is retrained. You can influence what retrieval finds, usually within weeks. What retrieval finds, for a product question, is mostly third-party pages: comparison articles, "best X" roundups, directories, review sites, Reddit and forum threads, and the vendors' own pricing and docs pages. The assistant then tends to recommend the products those pages mention repeatedly and describe consistently.
So the practical question is not "how do I talk to the AI". It is "am I on the pages it reads, and do those pages say the same thing about me".
Step 1: Can AI crawlers actually read your site?
Start here, because everything else is wasted if the door is locked. Each company runs separate bots for training and for search, and they are controlled separately in robots.txt.
| Bot | Company | What it does | Block it and you lose |
|---|---|---|---|
| Googlebot | Normal Search crawling, which also feeds AI Overviews and AI Mode | Google Search, AI features included | |
| Google-Extended | A control token for training and grounding in Google's other AI systems | Nothing in Search | |
| OAI-SearchBot | OpenAI | Finds pages to show in ChatGPT search | Appearing in ChatGPT search answers |
| GPTBot | OpenAI | Collects training data | Nothing in ChatGPT search |
| ChatGPT-User | OpenAI | Fetches a page when a user asks ChatGPT to read it | Live reads of your page on request |
| Claude-SearchBot | Anthropic | Indexes pages for Claude's search answers | Visibility in Claude's search answers |
| ClaudeBot | Anthropic | Collects training data | Nothing in Claude's search answers |
| PerplexityBot | Perplexity | Builds Perplexity's search index | Appearing in Perplexity answers |
| Bingbot | Microsoft | Bing crawling, which feeds Copilot and Bing's AI summaries | Bing and Copilot |
The table reflects each company's own documentation as of 12 October 2026. Two points are worth taking from it. You can block training and keep search: blocking GPTBot does not remove you from ChatGPT search, OpenAI's documentation says the two settings are independent, and changes take about 24 hours to be picked up. And Google has no separate AI crawler for Search, so if Googlebot can read you, AI Overviews can too.
Then check the layer most people forget, the CDN. Cloudflare began blocking AI crawlers by default on new domains in 2025, and plenty of founders have a robots.txt that welcomes every bot while their firewall turns them away. Open your CDN's bot settings and look for an AI crawler section. Your server logs will settle it either way: search them for OAI-SearchBot and PerplexityBot. If neither has visited in a month, find out why before doing anything else on this list.
Step 2: Do you describe your product the same way everywhere?
Assistants recommend what they can state confidently, and confidence comes from agreement. If your homepage calls you "an AI-powered growth workspace", your Product Hunt page says "a CRM for agencies" and your G2 profile says "client portal software", the model has three half-pictures and will usually pick a competitor whose story is simple.
Write one plain sentence that names the category, who it is for and the one thing that separates you: "Acme is invoicing software for freelancers that chases late payments automatically." Then use it, word for word, on your homepage, your pricing page, every directory listing, your social bios and your founder profiles. Keep a short version and a two-paragraph version beside it so nobody improvises. If you are not sure which sentence is true, the language your customers use about you is the best source, and a tool like PMF Finder turns support emails and reviews into positioning you can test.
The category word matters more than the clever part. People ask assistants for "invoicing software", not for "revenue operations reimagined". If the category is missing from your sentence, you are missing from the answer.
Step 3: Do your own pages answer buying questions plainly?
Your site is where assistants check facts: what you cost, who you are for, what you integrate with, what you do not do. Most SaaS sites hide those facts in images, toggles and sales-call language. Make them easy to lift.
- Put the answer first. Open each key page with a sentence that would make sense quoted on its own.
- Publish pricing as text. A pricing table rendered as an image or behind a "contact us" button gives the assistant nothing to repeat, so it repeats someone else's guess.
- Write the comparison pages yourself. "You vs your biggest competitor" and "alternatives to your biggest competitor" are the exact shapes of buying questions. Be fair in them, because a one-sided page reads like an advert and gets treated like one.
- Use specifics. Numbers, limits, dates and named integrations survive being summarised. Adjectives do not.
- Keep markup honest. FAQ and product schema help machines read the page, but Google's guidance says no special AI markup is needed and that structured data must match the visible text.
There is some research behind this. The 2024 Princeton-led GEO study tested changes against a simulated AI search engine and found that adding quotations, statistics and cited sources raised a page's share of the generated answer by roughly 28 to 41 percent, while keyword stuffing did little or harm. It was a lab setup, not live ChatGPT, so treat it as direction rather than a promise. To check the basics on a page, an on-page audit such as SEO Checker flags missing headings, meta tags and schema, and Kitset SEO Launch Kit drafts titles, robots rules and starter schema for a new site.
Step 4: Are you on the pages AI already reads?
This is the step with the biggest effect and the one most guides rush past. Pick ten buying questions your customers would ask, put them to ChatGPT, Perplexity and Google, and click every source cited. You will see the same few kinds of page over and over: roundup articles, comparison posts, directories, review sites and community threads. Those are your targets.
Directories and launch platforms. Curated directories are cited because they are structured: a category, a description, pricing and alternatives, all in one place. A listing only helps if it carries your one-sentence description and real details, so fill every field. Our guides to where to submit your SaaS and Product Hunt alternatives cover which ones are worth the time, and if you would rather hand off the form-filling, ListingBott does it with a person checking each submission.
Roundups and comparison articles. Find the "best X" articles the assistants cite for your category and email the authors with accurate, specific information: pricing, a screenshot, what you do differently, a fair note on who should not use you. Writers update these pages, and one good placement can feed answers for months.
Communities. Reddit and niche forums turn up constantly in AI citations. The only approach that lasts is answering questions well where your product is genuinely relevant, as yourself, with your affiliation stated. A listening tool like WhatsTrending helps you find those conversations while they are live. Planted praise gets spotted by moderators and often by readers, and the thread then says something worse about you than silence would.
Step 5: Can search engines index the pages that matter?
AI Overviews and AI Mode only use pages that are indexed and eligible to show a snippet in Google, and Copilot works from Bing's index. If your comparison page sits in "Crawled, currently not indexed", no amount of GEO will surface it. Check both Google Search Console and Bing Webmaster Tools, submit a sitemap to each, and fix whatever is keeping your key pages out. SitemapFixer diagnoses that particular Search Console status, and a broken link checker catches the dead internal links that strand pages.
How do you measure AI visibility?
Nothing measures it perfectly yet, so combine a few signals.
- A prompt panel. Keep a spreadsheet of 20 buying questions. Once a month, ask each one in ChatGPT, Perplexity, Claude, Gemini and Google AI Mode, and record whether you are mentioned, in what position and which sources were cited. It is manual and it is the most honest number you will get.
- Google Search Console. Since June 2026 it has generative AI performance reports showing impressions in AI Overviews and AI Mode. They show impressions, not clicks.
- Bing Webmaster Tools. Its AI Performance report, in preview since February 2026, shows how often Copilot and Bing's AI summaries cite your pages and for which queries. It does not cover ChatGPT.
- Category indexes and trackers. SERPRankings publishes a free index of which brands ChatGPT ranks across software categories, which is a quick way to see who owns your category today. Paid trackers from the large SEO suites automate the prompt panel at scale.
What doesn't work?
- llms.txt. No major AI provider has confirmed using it, and studies comparing sites with and without one found no difference in citations. We publish one ourselves, and we would not claim it has done anything measurable.
- AI-only markup or files. Google states plainly that AI Overviews need no special files or schema.
- Hidden instructions. Text aimed at AI crawlers that humans cannot see is a manipulation attempt, and search engines treat it as one.
- Self-ranking listicles. A "top 10" on your own blog with you at number one is easy for an assistant to discount, and it costs you credibility with human readers.
- Paying for placement inside answers. Where ChatGPT shows ads, they are labelled and kept apart from the answer. There is no way to buy your way into the answer itself.
Frequently asked questions about getting recommended by ChatGPT
What is generative engine optimization? Generative engine optimization, or GEO, is the work of making your product appear accurately in answers written by AI assistants such as ChatGPT, Perplexity, Claude and Google's AI Overviews. In practice it is SEO plus deliberate work on third-party mentions and consistent descriptions.
How long does it take to show up in ChatGPT answers? For search-based answers, usually weeks after your pages are indexed and cited by sources the assistant trusts. What a model remembers from training only changes when the model is retrained, which you cannot schedule.
Should I block GPTBot? Only if you object to your content being used for training. Blocking GPTBot does not remove you from ChatGPT search, which uses OAI-SearchBot. Block OAI-SearchBot and you disappear from ChatGPT search answers.
Does llms.txt help AI visibility? There is no evidence it does. No major assistant has confirmed reading it, and the studies so far found no lift in citations.
How can I check whether ChatGPT recommends my product? Ask it the questions your buyers would ask, in a fresh chat with search on, and note whether you appear and which sources it cites. Repeat monthly with the same questions so you are comparing like with like.
Where to go next
The fastest single step is usually Step 4. Start with the directories in where to submit your SaaS in 2026, check each one with the ninety-second test, and use the launch checklist so every listing carries the same description.
And if you have built something real, submit it to StackLedge. A person reads every listing, and each one carries a category, pricing and a plain description, which is exactly the kind of page assistants read when they build their lists.