How to rank on ChatGPT

1. How does ChatGPT actually pick sources?

When you ask ChatGPT something that needs current information, it doesn't answer from a single lookup. It runs a query fan-out: it rewrites your prompt into several related searches, pulls back candidate pages, and then synthesises an answer from the ones it trusts. Independent measurement of browsing behaviour puts this at roughly 1.4–2 searches per prompt on average — so the model is almost never ranking pages for the exact words you typed. It's ranking them for the sub-queries it invented from your prompt, which is the single most important thing to internalise.

Two consequences follow. First, you're not competing for one keyword — you're competing to be a consistent presence across a fan of related queries, so covering a topic thoroughly beats optimising a single page for a single phrase. Action: take the prompt you want to win, ask ChatGPT (or Perplexity) the same thing three times, note the actual sub-questions and sources it pulls, and make sure you have a crawlable, answer-first page for each of those sub-questions. Second, because the model blends several sources into one answer, being mentioned in the supporting cast — cited alongside other options — is more achievable and more durable than being the one canonical result. Aim to be reliably present, not uniquely dominant.

2. Make sure the AI crawlers aren't blocked

This is the most common own-goal, and it's invisible until you check. ChatGPT relies on OpenAI's crawlers — chiefly GPTBot (training/fetching) and OAI-SearchBot (search) — to read your pages. If your robots.txt disallows them, or a firewall or bot-management rule silently blocks them, you are structurally invisible no matter how good your content is. Plenty of sites block these bots by default or inherit a block from a security vendor without realising it.

Action: open https://yourdomain.com/robots.txt and search for User-agent: GPTBot, User-agent: OAI-SearchBot and User-agent: *blocks; remove any Disallow: / you didn't intend. Then curl your own page as the bot to confirm your CDN or WAF isn't returning a 403 — curl -A "OAI-SearchBot" -I https://yourdomain.com/ should return 200, not 403 or 503. Do the same for ChatGPT-User(the agent used when a user's ChatGPT session fetches a page live). Unblocking is the single highest-leverage fix available, because it's binary: blocked means zero, unblocked means eligible.

3. Server-render your content — AI crawlers don't run JavaScript

Even when they can reach your page, AI crawlers generally fetch the raw HTML and do not execute JavaScript the way a modern browser does. If your key content — product descriptions, answers, prices, the actual words — only appears after client-side rendering, the crawler sees an empty shell. This is the second most common reason a technically “live” page never gets cited.

Use server-side rendering or static generation so the meaningful text is present in the initial HTML response. Action: run curl -s https://yourdomain.com/your-page | grep "a sentence you want quoted" — if the sentence isn't in the raw response, the crawler can't see it either. Do this for prices, product descriptions and answers specifically, since those are the words you want lifted. If they only appear after client-side rendering, no downstream GEO tactic will help.

4. Structure content answer-first

ChatGPT is trying to extract a clean, quotable answer. Pages that lead with the answer are far easier to lift than pages that bury it under three paragraphs of preamble. Put the direct answer in the first one to two sentences under a heading phrased the way a person asks the question, then expand with detail, caveats and evidence below it.

Practically: use question-shaped H2s (“How long does it take to…”, “Which is cheaper…”), keep each answer self-contained so it makes sense pulled out of context, and prefer short declarative sentences over hedged ones. A well-built FAQ section is unusually effective here because each Q&A is already a standalone, extractable unit — which is exactly the shape a model wants.

5. Get mentioned on third-party sources and Reddit

This is the part most people underinvest in, and it's the one with the strongest evidence behind it. Ahrefs' study of about 75,000 brands found that the volume of brand mentions across the web correlated with AI visibility at 0.664, while classic backlinks correlated at only 0.218 — brand mentions were about three times as strongly associated. ChatGPT reads text; being named in the right context on a trusted page counts even when there's no link at all. Action: search yourbrand + "best [category]" and list every roundup you're missing from, then pitch the ones with a real editorial process.

Two source types punch above their weight. High-authority publications and category roundups — the “best X for Y” listicles and comparison articles people ask ChatGPT to summarise — are frequently pulled into answers, so getting fairly included in them is high-value. And community platforms, especially Reddit, which is the #1 cited source on Perplexity and a top source across engines; genuine, helpful participation in the threads where your buyers already ask questions is one of the most reliable ways to show up. Action: find the two or three subreddits where your category is discussed and answer real questions usefully under your own name. The rule is earn it, don't spam it — planted or astroturfed mentions get filtered and can poison your reputation with the exact communities you need.

6. Statistics, quotes and citations lift inclusion

How you write a page measurably changes how often it gets cited. The Princeton-led GEO study (KDD 2024) ran controlled experiments on what makes generative engines include a source, and found that adding statistics lifted a page's visibility in AI answers by roughly +37–41%, while adding direct quotations and named citations lifted it by about +28–40%, depending on the query type. Notably, keyword stuffing did not help — the model rewards credibility signals, not density.

So make your pages quotable: cite specific numbers with their source, attribute claims to named studies or people, and include a real data point rather than a vague adjective. “Brand mentions correlated at 0.66 vs 0.22 for backlinks (Ahrefs, ~75,000 brands)” is the kind of sentence a model happily lifts; “brand mentions are really important” is not.

7. Freshness and keeping pages current

AI answers lean toward content that looks current, especially for anything time-sensitive. A visible last-updated date, refreshed figures, and a page that reflects the present year signal that you're a maintained source rather than a stale one. This doesn't mean churning out new URLs — it means genuinely revising your best pages, correcting outdated numbers, and re-stating the date when you do.

Freshness compounds with everything above: a crawlable, answer-first page that you keep accurate will get re-fetched and re-cited, while a page you publish once and abandon slowly falls out of the answer set as the topic moves on around it.

8. Measure it honestly

You can't improve what you can't see, but AI visibility is noisy — re-asking the same question can surface a very different set of sources. Profound's tracking found that 40–90% of the domains an engine cites change over time, so a single check tells you almost nothing. Measure with a fixed set of prompts you care about, run them on a schedule, and look at the trend, not the day.

Two practical rules. Use confidence intervals, not point estimates, so you can distinguish a real movement from sampling noise before you act on it. And you only need to check once a day — sampling more often mostly adds noise and cost without adding signal, because the underlying answer distribution doesn't change meaningfully hour to hour.

9. Common mistakes that keep you out of ChatGPT

Most brands that can't get cited are tripping over one of a handful of avoidable errors. Work down this list before you invest in anything more exotic:

• Assuming Google rankings carry over. They largely don't: across Ahrefs' analysis of ~863,000 SERPs, only about 37.9% of the URLs AI engines cite also rank in Google's top 10. Ranking #1 on Google is neither necessary nor sufficient — GEO is its own game.
• Silently blocking the crawlers. A security vendor's default rule, an over-broad Disallow: /, or a 403 to unknown user-agents makes you invisible with zero warning.
• Client-side-only content. If the words render in the browser but not in the raw HTML, the crawler sees an empty shell.
• Optimising for the prompt, not the fan-out. Targeting the exact phrase while ignoring the 1.4–2 sub-queries ChatGPT actually searches leaves you absent from the searches that decide the answer.
• Vague, unquotable copy. “We're the leading solution” gives the model nothing to lift; a specific number with a named source does.
• Talking only about yourself. With brand mentions correlating at 0.664 vs 0.218 for backlinks, the reputation lives off-site — no amount of homepage copy substitutes for being named on pages you don't own.
• Judging by one check. With 40–90% of cited domains rotating over time, a single query will mislead you in both directions.

How does steek help?

steek.ai is built to make this measurable without guesswork. It tracks how often ChatGPT and the other AI engines mention and recommend your brand across a prompt set you choose, puts a confidence interval on every number so you can tell signal from noise, and shows which third-party sources the engines are pulling from — so you know where to earn your next mention. If you want to go deeper, see our roundup of the best AI visibility tools, the honest steek vs Peec comparison, or just run the free live checker on the homepage — no signup — to see where you stand on ChatGPT right now.