LLM SEO — How ChatGPT, Gemini and Perplexity Pick Their Sources

ChatGPT chooses sources by running live web searches, pulling the pages that best match the sub-questions behind your prompt, and citing the passages it used to write the answer. It relies on search partners like Bing plus OpenAI's own crawler, so a page has to be crawlable, indexed and clearly answer the question to get cited.

Key takeaways

• LLM SEO is the work of getting your pages retrieved and cited by AI answer engines. Retrieval comes first, so classic indexing and ranking still decide most of it.
• ChatGPT leans on Bing and OAI-SearchBot, Gemini and AI Overviews use Google's index, and Perplexity runs its own crawler.
• Engines cite passages, not whole pages. Self-contained answers, lists, tables and sourced numbers get lifted most.
• Reddit is cited heavily because it's full of first-hand, question-shaped discussion. Brands need to be present there honestly, not spam it.

LLM SEO (also called SEO for AI or AI search optimization) is about one question: when someone asks ChatGPT, Gemini or Perplexity something in your niche, does the answer mention you and link to you? This guide explains the mechanics behind that answer and gives you a framework you can run on your own site.

If you're still sorting out the terminology, my GEO vs SEO breakdown covers where the two overlap. Here I'm going deeper on the machinery.

How LLM search works (retrieval + generation)

A language model on its own only knows what was in its training data. That data has a cut-off date and no reliable link back to sources. So every AI search product bolts a search system onto the model. The industry name for this is retrieval-augmented generation, or RAG.

The flow looks roughly like this:

1. The model reads the prompt and decides whether it needs fresh information. A question like "write me a poem" doesn't trigger search. "Best CRM for a 10-person agency" usually does.
2. It rewrites the prompt into search queries. One prompt often becomes several queries. Google calls this "query fan-out" and says AI Overviews and AI Mode issue "multiple related searches across subtopics and data sources" to build a response (Google Search Central).
3. A search index returns candidate pages. This is plain old ranking. If you don't rank somewhere for those sub-queries, you're not in the candidate pool.
4. The system pulls passages from those pages and hands them to the model as context.
5. The model writes the answer and attaches citations to the passages it leaned on.

Google's Gemini API documentation describes the same loop for its "Grounding with Google Search" feature: the model decides whether search would help, generates one or more queries, processes the results and returns an answer with citations mapped to specific text segments.

Two things follow from this.

First, steps 2 and 3 mean LLM SEO sits on top of normal SEO. A page that isn't indexed, or ranks on page 9 for every sub-query, rarely gets pulled.

Second, step 4 means the unit of competition is the passage. Your page can be long and excellent, but if no single paragraph cleanly answers a sub-question, the model will quote someone else's paragraph.

The model's training data still matters, but differently. It shapes which brands the model already "knows" and how it phrases its search queries. A brand that shows up consistently across the web gets named more readily, even before retrieval kicks in. That's the slow, reputation side of LLM SEO. Retrieval is the faster, page-level side.

Where each engine gets its sources (Bing, Google, own index)

Each engine uses a different index, and that's why your citations can look completely different across them for the same prompt.

ChatGPT. OpenAI's help centre says ChatGPT search "sometimes partners with other search providers" and names Microsoft Bing among them. It also says that to be eligible you should allow OAI-SearchBot to crawl your site (OpenAI Help Center). OpenAI's crawler documentation is blunt: "Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers." GPTBot is a separate crawler used for model training, so you can block training and still allow search.

Gemini and Google AI Overviews / AI Mode. These draw on Google's own index. Google states there are "no additional requirements to appear in AI Overviews or AI Mode", and a page just needs to be indexed and eligible to show with a snippet. The Google-Extended robots.txt token controls whether your content is used for Gemini training and for grounding in Gemini Apps and Vertex AI, but Google says it "does not impact a site's inclusion in Google Search".

Perplexity. Perplexity runs its own crawler. Its docs say PerplexityBot is "designed to surface and link websites in search results on Perplexity", while Perplexity-User fetches pages live when a user asks a question. My Perplexity SEO guide goes through its quirks in detail.

Microsoft Copilot. Runs on Bing. Bing Webmaster Tools now has an AI Performance report (public preview from February 2026) that shows citation counts and the "grounding queries" behind them, which is the closest thing to first-party AI citation data any engine offers right now.

The practical takeaway: if you've only ever checked Google Search Console, go verify your site in Bing Webmaster Tools today. ChatGPT and Copilot both lean on Bing, so skipping it means ignoring two of the five engines above.

What gets cited: passages, lists, stats and entity-clear brands

Nobody outside these companies has the ranking code. What we do have is Google's documentation, the engines' own help pages, and large third-party citation studies. Together they point to a consistent set of traits.

Self-contained passages

A citation points at a chunk of text. The chunk that wins usually answers the sub-question in its first sentence and doesn't depend on the paragraph above it. "It depends, as mentioned earlier" is useless out of context. "Most small sites see first rankings in 3 to 6 months" is quotable.

This is why every AlgoBlueprints post opens with a 40–60 word answer paragraph. I've written more on how this works for Google specifically in how AI Overviews pick citations.

Lists and tables

Comparison prompts ("X vs Y", "best tools for Z") produce answers built from lists. Pages that already present options in a scannable list or table are easy for a model to map onto its answer structure.

Specific, sourced numbers

Models like to back claims with figures. A page that states a number, says where it came from and when, gives the model something concrete to cite. Vague claims ("prices vary widely") give it nothing.

Entity-clear brands

When a model decides which brands to mention, it's pattern-matching across everything it has seen. A brand that describes itself the same way on its site, LinkedIn, directories, reviews and press is easier to place in a category than one with a fuzzy, shifting description. Think of it as consistency of the "who we are and what we do" sentence across the web.

Things that don't seem to matter much

Google says you don't need AI text files or special markup for AI Overviews. That includes llms.txt, which I've covered in my llms.txt explainer. It's cheap to add but no major engine has confirmed using it for ranking or citation.

The role of Reddit and forums in AI answers

Reddit shows up in AI answers constantly, and the numbers back that up.

Profound analysed 680 million citations from August 2024 to June 2025. Reddit was the top cited domain in both Google AI Overviews (2.2% of citations) and Perplexity (6.6%). In ChatGPT, Wikipedia led at 7.8%, with Reddit second at 1.8% (Profound).

A Semrush study of 230,000+ prompts between July and October 2025 found something more volatile: Reddit appeared in close to 60% of ChatGPT responses in early August, then dropped to around 10% by mid-September, while staying steady on Perplexity and AI Mode (Semrush).

That swing is the most useful lesson here. Citation patterns can change overnight when an engine adjusts its retrieval. Don't build your whole strategy on one domain.

Why Reddit gets cited

• It matches how people prompt. AI prompts are long, conversational questions. Reddit threads are long, conversational answers to exactly those questions.
• First-hand experience. "I used this tool for 8 months, here's what broke" is the kind of detail marketing pages avoid and models find useful for balanced answers.
• Freshness and volume. New threads appear on almost every topic every day.
• Data access. Reddit signed data licensing agreements with Google and OpenAI in 2024, which gives both companies structured access to its content.

What it means for brands

You can't fake your way into Reddit citations. Moderators and users spot astroturfing fast, and a thread calling out your brand as spam is also citable.

What works is slower:

• Search your brand and category on Reddit to see what's being said. Those threads may already be feeding AI answers about you.
• Have a real person (founder, support lead) answer questions in relevant subreddits, disclosing who they are.
• Fix the product problems people complain about. A model summarising "common complaints about X" will find them.
• Publish the kind of first-hand detail Reddit rewards on your own site too, so you're a candidate source either way.

Quora, niche forums, Stack Exchange and YouTube follow the same pattern at smaller scale.

The 8-step LLM SEO framework

This is the order I'd work in for a site that has decent content but low AI visibility. Steps 1–3 are about getting into the candidate pool. Steps 4–8 are about winning the citation once you're there.

1. Check crawler access. Open robots.txt and make sure you aren't blocking Googlebot, Bingbot, OAI-SearchBot or PerplexityBot. Check your CDN or firewall too, since bot-protection rules often block AI crawlers without anyone noticing. Blocking GPTBot (training) is a separate decision and doesn't affect ChatGPT search.
2. Get indexed in both Google and Bing. Verify in Google Search Console and Bing Webmaster Tools, submit your sitemap to both, and turn on IndexNow if your CMS supports it. If pages aren't indexed, start with my guide on why a website isn't showing on Google.
3. Map the fan-out queries. For each target prompt, list the sub-questions an engine would search. "Best SEO tool for a small agency" fans out into pricing, features, reviews, alternatives and comparisons. Make sure you have content that ranks for those pieces.
4. Write answer-first passages. Under each H2, open with one or two sentences that answer the heading fully on their own. Then add detail.
5. Add structure models can reuse. Comparison tables, numbered steps, pros and cons, definitions. Match the shape of the answer you want to appear in.
6. Back claims with sourced specifics. Numbers with a source and date, named tools, real prices, real limits. Mark anything you can't source and fix it before publishing.
7. Make your brand entity consistent. Same name, same one-line description, same category on your site, About page, social profiles, directories and author bios. Use Organization schema to state it on-site.
8. Earn mentions off-site, then measure. Get cited in industry roundups, podcasts, Reddit discussions and comparison articles. Then track a fixed prompt set monthly. My AI search visibility guide explains the measuring side.

None of these steps is exotic. That's the point. Most of LLM SEO is SEO done with passage-level discipline and a wider view of where your brand gets mentioned.

Run your own citation test

To see who gets cited in your niche, run 10 prompts your readers actually ask in ChatGPT and Perplexity and record which domains are cited. Use a logged-out or fresh session, write the prompts the way a customer would phrase them, and record the date. Results shift week to week, so a single run is a snapshot, not a verdict.

Where LLM SEO fits with the rest of your SEO

I treat LLM SEO as a layer on top of an existing strategy. The page that ranks in Google is the page that gets retrieved and cited; the extra work is passage formatting, Bing coverage, crawler access and off-site mentions.

For ChatGPT specifically, my guide to ranking in ChatGPT goes through prompt testing and Bing-side fixes step by step. For Google's AI features, see how to rank in AI Overviews.

Want to see where you stand right now? Pick 10 prompts, run them this week, and use my AI search visibility tracking method to log the results.

Frequently asked questions

How does ChatGPT choose sources?

ChatGPT turns your prompt into search queries, sends them to search partners such as Bing and its own OAI-SearchBot index, then pulls passages from the best-matching pages. It cites the passages it used. Pages must be crawlable by OAI-SearchBot, indexed, and clearly answer the sub-question to be picked.

What is SEO visibility?

SEO visibility is a measure of how often your site appears in search results for a set of tracked keywords, usually weighted by ranking position and search volume. AI visibility applies the same idea to AI answers: how often your brand is mentioned or cited across a fixed set of prompts.

Is LLM SEO different from regular SEO?

Mostly no. AI engines retrieve pages through search indexes, so indexing, relevance and authority still decide who gets considered. The differences are emphasis: answer-first passages, Bing coverage, allowing AI search crawlers, and building brand mentions on forums and third-party sites that models read.

Does blocking GPTBot remove my site from ChatGPT search?

No. GPTBot collects data for model training. ChatGPT search uses OAI-SearchBot, and OpenAI says sites that block OAI-SearchBot won't appear in ChatGPT search answers. You can block GPTBot and still allow OAI-SearchBot if you want search visibility without training use.

Why does Reddit appear so often in AI answers?

Reddit threads match how people prompt AI: long, specific questions answered with first-hand experience. Citation studies from Profound and Semrush found Reddit among the most-cited domains in Perplexity, Google AI Overviews and ChatGPT, though its share in ChatGPT dropped sharply in September 2025.

A

Akif Qureshi

Senior SEO Specialist & Marketing Analyst | Content Strategist

5+ yrs experience Google Certified 6 guides

About

Driven by advanced SEO expertise, deep marketing analytics, high-impact content strategy

With 5+ years of hands-on experience, I specialize in holistic search strategies that don’t just rank—they drive real, measurable business growth. I’ve worked across industries including healthcare, hospitality, legal, e-commerce, and professional services, helping brands dominate their target markets. My approach bridges the gap between raw data and creative execution. Every strategy I build is rooted in rigorous market analysis, structured SEO frameworks, and tailored content ecosystems—no templates, no shortcuts. Whether you’re a single-location brand or scaling across multiple cities, I create data-driven marketing systems designed to compound results and grow with you.

No sponsored content No affiliate links Reader supported

Want Your Brand Cited in AI Answers?

Get an AI search visibility audit covering ChatGPT, Perplexity and Google AI Overviews — and a plan to earn more citations.

Book a Free SEO Audit →