You cannot rank in ChatGPT. There is no index and no position three. When someone asks it "what's the best invoicing tool for freelancers?", it does not consult a league table. It writes a paragraph, and that paragraph names a few products. Either you are one of them, or for that buyer you do not exist.
So the honest version of "how to rank in ChatGPT" is a different question: how do I become one of the names the answer reaches for? That one has real, boring, doable answers. It also has a pile of superstition around it that will happily eat months of your marketing effort. Here is how I would spend the effort, in order, and what I would skip.
Where the names in an answer come from
Every ChatGPT answer is fed by two supply lines, and they have almost nothing in common:
| The trained model | The live search step | |
|---|---|---|
| What it is | Knowledge baked in during training | Web pages fetched while answering |
| How current | Frozen at the training cutoff | Today |
| Who gets named | Brands the public record talked about, consistently, for years | Brands named in the handful of pages it just read |
| How fast you can influence it | Months to years | Weeks |
| Your lever | Consistency everywhere | Presence in the pages it retrieves |
Every tactic that works feeds one of these two lines; the rest is an attempt to skip them. If you want the full mechanics of how the two supply lines behave, Lasha has written a deeper explainer; this post is the to-do list.
Door one: the pages ChatGPT reads when it searches
When the question sounds like a buying decision — "best", "vs", "for small teams", "in Berlin" — ChatGPT will often run a web search before answering. It reads a handful of pages and builds the shortlist mostly out of what those pages say. Sit with that for a second, because it is the whole game: the answer borrows its names from other people's pages.
Which pages? You don't have to guess. Ask ChatGPT your buyer's question and read the citations at the bottom of the answer. Those URLs are the actual distribution channel for your category — usually a mix of "best X for Y" roundups, review platforms, comparison articles, and Reddit threads. Then do the obvious, unglamorous work:
1. Get named in the cited pages. If the same three roundups keep deciding every answer in your category, being added to those three articles beats publishing thirty of your own. Pitch the authors. Get listed on the review platforms they lean on. Show up in the forum threads with something useful to say. This is PR and community work, not a technical trick, and it is the single biggest lever you have.
2. Check your plumbing. ChatGPT's search results come through OpenAI's own crawler, OAI-SearchBot, and OpenAI is explicit that sites blocking it get dropped from ChatGPT search answers (at best you surface as a bare link). So: confirm your robots.txt allows OAI-SearchBot. While you're at it, verify your site in Bing Webmaster Tools — OpenAI named Bing as part of the search mix at launch and has never announced dropping it, so a clean Bing presence is cheap insurance. This is an hour of work, once, and the bar is binary: not blocked.
3. Publish the answer-shaped page. Write the page that directly answers your buyer's exact question, with the comparison table and the straight recommendation a lazy reader needs. Best case, ChatGPT starts citing you. Worst case, you have a good landing page.
Door two: what the model already believes
Ask an evergreen question — "what are the well-known CRM tools?" — and ChatGPT frequently answers without searching at all, from its training data. No page you publish this month touches that answer. What feeds it is the public record about your brand, accumulated over years: your site, directories, press, reviews, forums, all of it compressed during training.
You influence this the slow way. Make the record agree about what you are: the same name, the same one-line description of what you do, everywhere it appears. Clean up the profile pages that describe you wrongly. Earn mentions in the places that describe your category. None of this pays off next week. It pays off across model versions, which is why it belongs in the plan but not at the top of it.
You can't force the search, either
One thing you do not control: whether ChatGPT searches at all. The model decides per question, and it decides less often than most advice assumes. When we set our own tracker to the setup a real free-tier user actually gets, instead of the developer API defaults, searches dropped to about half — Lasha's explainer has the details. A big chunk of the answers your buyers see never touch the web at all, which is exactly why door two exists.
The practical consequence: check your actual questions. If they trigger searches, door one pays fast. If they get answered from memory, you are in the slow game whether you like it or not, and you should know that before you budget the quarter.
The plan, in order
- Today: ask ChatGPT (and Gemini, and Claude) your five most important buyer questions, logged out. Save every answer and every cited URL.
- This week: the robots.txt / OAI-SearchBot check and Bing verification. One hour, done forever.
- This month: get your brand into two or three of the pages that kept showing up in those citations. This is the lever.
- Also this month: publish the answer-shaped page for your single most valuable question.
- Ongoing: make the public record consistent about what you are.
- Every week, starting now: measure on a schedule, because a single check tells you almost nothing — the same question produces different shortlists run to run, and you need a rate, not a screenshot. (How to do a clean manual check is its own post.)
What to skip
Betting real budget on llms.txt. Google has said it won't crawl or use it, and OpenAI's crawler documentation assigns it no role. Publishing one costs nothing, so publish one if you like. Budgeting for it is a different matter.
Magic markup. There is no schema.org type that gets you into AI answers. Google says this outright about its own AI features, and nobody has produced evidence to the contrary for ChatGPT. Use structured data for the rich results it actually earns.
"AI-optimizing" your existing copy. Rewriting pages to sound more machine-readable, without changing what the pages say or where your brand is mentioned, changes nothing that either supply line reads.
Anyone selling guaranteed placement. No assistant currently sells organic placement inside composed answers. A vendor who guarantees mentions is guaranteeing something they do not control.
How you'll know it worked
Pick your questions, run them on a schedule across the assistants, and watch the rate: how often you are named, where in the answer, alongside whom, and whether the model searched. That last field tells you which door is moving. If you want this automated — real consumer-tier questions, every answer scored the same way — that is what Mentionify does, and you can run one free check right now to see where you stand.
FAQ
How long does it take to show up in ChatGPT?
Through the search door: as soon as the pages it reads start naming you, which in practice means weeks — the time it takes to get added to a roundup or accumulate review-platform presence. Through the training door: model-version timescales. Anyone promising days is describing an ad.
Can I pay to be recommended?
No. You can buy ads around AI answers; nobody currently sells the sentence itself. Anyone offering guaranteed mentions is selling air.
Does this work the same for Gemini and Claude?
The two supply lines are the same shape everywhere: trained knowledge plus an optional live search. What differs is temperament — how readily each model searches and how many brands it names — which is why measuring each assistant separately matters more than optimizing for any single one.
Do I need a Wikipedia page?
It helps the slow door and it is famously hard to get legitimately. Consistent presence across review platforms, directories, and category articles is more achievable and feeds the same record.
