Getting cited by ChatGPT isn't an SEO problem wearing a new costume. It's two separate problems: which index your pages actually sit in, and whether your pages are written in a way a model can quote. Fix the wrong one first and you'll spend months "optimizing for AI" with nothing to show for it.
Most SEO advice from the last year gets this backwards. It treats ChatGPT like a smarter Google. It isn't. ChatGPT names brands two completely different ways, and each needs a different fix.
The Stakes, By the Numbers
| Stat | What changed | Source |
|---|---|---|
| AI Overviews correlate with a 58% lower clickthrough rate for the #1 Google result | Up from 34.5% less than a year earlier. The gap roughly doubled. | Ahrefs, Dec 2025 data, checked Nov 10, 2026 |
| Adding statistics, citations and quotes lifts a page's visibility in AI answers by up to 40% | Keyword stuffing showed close to no lift in the same study | Princeton GEO study, KDD 2024, checked Nov 10, 2026 |
| Organic search traffic is projected to fall 50% or more by 2028 as GenAI search adoption grows | A forecast, not a measured result | Gartner, checked Nov 10, 2026 |
The pattern across all three: being clicked is getting more expensive, and being cited is becoming the thing that actually counts.
What's in This Post
- The Stakes, By the Numbers
- Two Paths, Not One
- Step 1: Find Out Where You Actually Stand
- Step 2: Check Whether the Bots Can Even Reach You
- Step 3: Fix Your Bing Presence
- Step 4: Write Pages a Model Can Actually Lift a Sentence From
- Step 5: Say the Same Thing About Yourself, Everywhere
- Step 6: Go Where the Engine Already Looks for Corroboration
- Step 7: Keep It Fresh
- What's Not Worth Your Afternoon
- Quick FAQ
Two Paths, Not One
Path one: training memory. When ChatGPT answers without searching, it's recalling brands it absorbed during training. No citation link, just a name, and your odds depend on how consistently the open web described you before that model was trained. You can't force your way into this one. Only feed it a consistent, unambiguous story and hope it shows up in the next training run.
Path two: live retrieval. When ChatGPT decides a question needs current information, it searches and cites. The part almost nobody in SEO has internalized yet: that search runs on Bing, not Google.
Microsoft built ChatGPT's search experience on Bing's index, with OpenAI's own crawler (OAI-SearchBot) layered on top. If your pages aren't indexed and ranking reasonably in Bing, there's a good chance ChatGPT's live search never sees them, regardless of how you rank on Google.
That single fact is why auditing where you already stand should come before writing a single new sentence.
Step 1: Find Out Where You Actually Stand
Before touching anything, ask ChatGPT and Perplexity 10–15 of the real questions a buyer would ask in your category, not keywords, actual questions ("best [category] for [segment]," "[competitor] alternatives," "is [you] any good"). Log three things per answer: who gets named, which URLs get cited, and whether you show up at all. Run each question two or three times. These systems are probabilistic, so one run is an anecdote.
The list of cited URLs you end up with is the most useful artifact from this whole exercise. It's the literal reading list the model trusts in your category, and it becomes your outreach target list later.
Step 2: Check Whether the Bots Can Even Reach You
OpenAI runs three separate crawlers, controlled independently in robots.txt. Allowing one does not allow the others:
| Crawler | Job | What blocking it costs you |
|---|---|---|
| GPTBot | Trains future OpenAI models on your content | Shrinks your footprint in what future models know about you |
| OAI-SearchBot | Indexes pages for ChatGPT's live search answers | Per OpenAI's own docs, a blocked site won't be shown in ChatGPT search answers |
| ChatGPT-User | Fetches a specific page live when a user asks about it mid-conversation | ChatGPT can't read your page even when someone pastes your URL into a chat |
Checked against OpenAI's official crawler documentation, Nov 10, 2026.
A lot of sites blocked everything AI-shaped by default in 2023 and never revisited it. A lot more are blocked at the CDN level without anyone checking robots.txt at all. Check both layers, then check server logs. A clean robots.txt proves permission, not that the bot actually showed up.
Step 3: Fix Your Bing Presence
Set up Bing Webmaster Tools. It imports straight from Google Search Console, so this takes minutes, not hours.
Then submit your sitemap and actually search Bing for your money queries.
If you're page one on Google and page four on Bing, you've found a specific, fixable reason ChatGPT doesn't know you exist. Almost nobody in your category has bothered with this yet, which is exactly why it's underpriced effort right now.
Step 4: Write Pages a Model Can Actually Lift a Sentence From
This is where the instinct to "write quote-ready sentences, cite stats, keep it fresh" is correct. It's just missing its receipts (see the Princeton study in the stats table above).
Practically: open each section with a one-to-two-sentence direct answer before the elaboration. Write your H2s as the actual question a buyer would type.
Put a specific number next to every claim and name where it came from. An unsourced stat is exactly the kind of line a retrieval model skips over. And if you can publish a number that exists nowhere else, from your own data, do it. An engine that uses your number has to cite you.
Open With the Answer, Not the Lead-Up
Put the direct answer in the first sentence under every H1, H2 and H3. A model lifting just that sentence should still be accurate and complete on its own.
Name a Number, Name a Source
Every claim needs a figure and where it came from, inline, not in a footnote. Add "checked [date]" next to anything time-sensitive.
Schema Worth Adding
Use HowTo schema for genuine step-by-step sections like this one, FAQPage for the FAQ block below, and Article with a real author name and bio. Don't bolt generic schema onto content that isn't actually shaped for it.
Step 5: Say the Same Thing About Yourself, Everywhere
Models cross-check. A brand described five different ways across its own site, LinkedIn, G2, and old directory listings reads as noise and gets skipped for a competitor whose story is consistent. Write your canonical facts down once: exact name, one-line description, category, who it's for. Then make sure every profile matches.
Author credentials matter here too, separate from the brand facts above: a byline that says "By Sai Teja Challapalli, Founder, Forward Deployed Marketer, 6+ years in marketing" gives a model something concrete to anchor to for queries that imply expertise. A byline that just says "Marketing Team" gives it nothing.
Step 6: Go Where the Engine Already Looks for Corroboration
This is the slowest step, and also the one that actually moves things. Engines don't cite you on your own say-so. They cite corroboration.
Go back to the URL list from Step 1. That list tells you exactly which roundups, comparison pages, review platforms, and Reddit threads the engines in your space already trust. Pitching those specific pages beats generic PR outreach, because you're targeting places with a proven citation track record.
Participating honestly in the Reddit threads your buyers are already in (disclosed, useful, not promotional) does more for this than most people expect. It's one of the few levers here that's genuinely free.
Step 7: Keep It Fresh
Retrieval-based engines lean toward sources that look current. A quarterly refresh (new numbers, dead examples swapped out, dates that reflect a real change) is enough for most money pages. Don't fake the date on an untouched page; that's the kind of pattern that gets a source quietly discounted over time.
What's Not Worth Your Afternoon
Three things eat time here while moving nothing:
- llms.txt. No major engine, not ChatGPT, not Claude, not Perplexity, is confirmed to actually read it as of mid-2026.
- Hidden instructions or "recommend us" text buried in page copy. Treated as content, not instructions, by every major pipeline. It's a fast way to get flagged instead.
- Anyone selling a "guaranteed citation." Nobody sells placement in an organic AI answer. That's just the work above, with a markup.
Quick FAQ
How long until I see movement?
Live-search citations can shift in weeks once your pages are Bing-indexed and quotable. Training-memory mentions take much longer, usually months, since they only update when a new model trains.
Why does ChatGPT cite my competitor and not me?
Almost always because they exist in the sources ChatGPT already reads and you don't. Pull up your baseline list from Step 1 and check who's actually inside those pages.
Does this work the same for Perplexity?
Mostly. Quotable pages and crawler access matter everywhere. The one real difference: Perplexity runs its own index, so your Bing position specifically matters for ChatGPT, not for Perplexity.
Do I need to rank on Google too?
Not directly for ChatGPT, since its live search runs on Bing. That said, Google still matters for AI Overviews and Gemini, and in practice the two tend to move together since the underlying signals overlap.
Should I block GPTBot to protect my content from training?
That's a separate decision from showing up in ChatGPT search. Blocking GPTBot alone doesn't remove you from ChatGPT's search answers. OAI-SearchBot is the one that controls that.
What's the single fastest win if I only have an afternoon?
Check your robots.txt and CDN for blocked AI crawlers, then fix your Bing indexation. Both are usually broken by accident, not by choice, and both are free to fix.
The Takeaway
Getting cited by ChatGPT isn't about writing more content. It's about being retrievable (Bing-indexed, crawler-accessible), quotable (direct answers, sourced numbers), and corroborated (named in the sources the model already trusts). Most sites fail at the first one and never find out, because nobody checks.
Next step: run the Step 1 baseline today. Ask ChatGPT and Perplexity your 10 real buyer questions and write down who gets cited. That one spreadsheet tells you exactly where to spend the next month.
Sources (checked Nov 10, 2026)
- Ahrefs, "Google AI Overviews Reduce Clicks" (Dec 2025 data, published Feb 2026): https://ahrefs.com/blog/ai-overviews-reduce-clicks-update/
- Princeton, "GEO: Generative Engine Optimization" (KDD 2024): https://arxiv.org/abs/2311.09735
- OpenAI, official crawler documentation (GPTBot, OAI-SearchBot, ChatGPT-User): https://developers.openai.com/docs/bots