GEO prompts: get cited by ChatGPT & Perplexity.

How to use these (3 minutes)

Run prompts 1–5 on ChatGPT (with browsing on) and prompts 6–8 on Perplexity. Run each prompt 2–3 times — citations vary between runs, so average the results. Log which domains get cited in a simple sheet. Prompts 9–10 check your technical setup. Don’t trust a single run; AI citations are volatile.

01 · The entity test (ChatGPT)

Tests: whether ChatGPT even knows your brand exists — the absolute baseline.
Run on: ChatGPT, fresh chat, browsing on.

What is [YOUR BRAND NAME]? What does it do, and where can I find its work?

[Replace with your exact brand name. Run 3 times. Note: does it describe you correctly, confuse you with someone else, or say it doesn't know you?]

02 · The unprompted recommendation (ChatGPT)

Tests: whether ChatGPT recommends you without being asked — the query that maps to revenue.
Run on: ChatGPT, fresh chat.

Who should I hire for [YOUR SERVICE] in [YOUR CITY/AREA]?

[Example: "Who should I hire for SEO in Ghaziabad?" Run 3 times. Log every brand cited. Are you on the list? Who beats you?]

03 · The best-X list test (ChatGPT)

Tests: listicle presence — studies report ~44% of ChatGPT-cited pages are "best X" lists.
Run on: ChatGPT, fresh chat.

What are the best [YOUR CATEGORY] options right now?

[Example: "What are the best SEO agencies for local businesses in India?" Note which listicles get cited — then check: are YOU featured in any of those lists? That's your fastest win.]

04 · The comparison test (ChatGPT)

Tests: whether ChatGPT has enough material to represent you fairly against a competitor.
Run on: ChatGPT, fresh chat.

Compare [YOUR BRAND] with [NAMED COMPETITOR]. Which is better for [SPECIFIC NEED]?

[If it can't compare you, or invents facts about you, your entity footprint is too thin. Note every invented claim — that's what prospects see.]

05 · The freshness test (ChatGPT)

Tests: whether your content is current enough — studies report 76–90% of cited pages were recently updated.
Run on: ChatGPT, fresh chat.

Is [A SPECIFIC CLAIM FROM YOUR SITE] still accurate as of today?

[Pick a claim with a date or number, e.g. pricing. If it quotes outdated info, your page needs a refresh with a visible updated date.]

06 · The entity test (Perplexity)

Tests: the same baseline as prompt 01 — but Perplexity retrieves live, so results often differ completely.
Run on: Perplexity, new thread.

What is [YOUR BRAND NAME]? What does it do?

[Compare with prompt 01's answers. Different sources cited? That's normal — the two engines share only ~11% of cited domains. You need both.]

07 · The live-answer test (Perplexity)

Tests: Perplexity's live retrieval — it favors fresh, community, and primary sources over encyclopedic ones.
Run on: Perplexity, new thread.

What is the best approach for [YOUR AUDIENCE'S COMMON QUESTION] in 2026?

[Example: "What is the best approach for local SEO for a new business in 2026?" Perplexity cites a median of ~6 sources per answer vs ChatGPT's ~3 — more slots, more chances. Log all cited domains.]

08 · The question-heading test (Perplexity)

Tests: whether your pages are structured to be quoted — question-based headers get extracted most.
Run on: Perplexity, new thread.

[PASTE ONE PAGE'S MAIN CONTENT, TEXT ONLY]

If someone asked "[THE QUESTION THIS PAGE ANSWERS]", which exact sentences from this text would you quote? What would you have to look up elsewhere?

[Whatever it "looks up elsewhere" is your content gap. Add it to the page under a question-based heading.]

09 · The crawler audit

Tests: whether your robots.txt blocks the wrong AI bots — the most common technical GEO mistake.
Run on: any AI chat.

Here is my robots.txt:

[PASTE ROBOTS.TXT]

Which AI crawler user-agents does this block or allow? Specifically: am I blocking OAI-SearchBot, ChatGPT-User, PerplexityBot, or ClaudeBot (the retrieval bots that power live citations)? Am I blocking GPTBot or Google-Extended (training bots — blocking those is fine and doesn't hurt citations)? Give me the corrected robots.txt.

10 · The citation tracker setup

Tests: nothing directly — it builds your ongoing measurement system.
Run on: any AI chat.

I want to track my AI citation share weekly. My business: [WHAT YOU DO], audience: [WHO], area: [WHERE].

Build me a table of 20 prompts my real customers would ask (mix: "best X", "how do I", "X vs Y", pricing questions). For each: which engine to run it on (ChatGPT/Perplexity), and columns to log: date, run 1 citations, run 2 citations, my brand cited? (y/n). Include a scoring rule: brand cited in 2+ runs = 1 point.

What the evidence actually says

Most GEO guides recycle the same advice. Here’s what published research and large-scale studies actually report — with the caveats the guides skip:

• The Princeton paper is real, but read it carefully. Aggarwal et al.’s “GEO: Generative Engine Optimization” (arXiv:2311.09735, KDD 2024) found up to 40% visibility lift from tactics like citing sources, adding quotations, and including statistics (30–40% band). But it was tested on a custom-built engine, not production ChatGPT or Perplexity.
• A 2026 replication on 3,205 real pages tells a harder story. Statistics density replicated on all four production platforms — but citation and quotation density went the wrong direction. Hypothesis: engines cite primary sources, not pages that merely aggregate citations. (Source: AI+Automation, March 2026 data.)
• ChatGPT ≠ Perplexity. Test both. Studies report the two share only ~11% of cited domains. ChatGPT retrieves largely via the Bing index (only ~30% of its citations come from Google’s top-10); Perplexity uses live retrieval (~89% from Google’s top-10). One “GEO strategy” is provably wrong.
• Freshness beats almost everything. 76–90% of cited pages were recently updated. Word count correlates ~zero with citations. Schema markup showed ~zero measurable citation impact in Ahrefs’ large-scale analysis.
• “Best X” lists dominate ChatGPT citations — ~44% of cited page types in one large study. If your category has listicles and you’re not on them, that’s the gap.
• India-specific: a 2026 India study found Gemini cites 90% of English answers but only 41.3% of Hinglish ones. If your audience searches in Hinglish, English-first content gets cited more.

Where most GEO guides lie to you

• “Add llms.txt and AI will cite you.” llms.txt is a real proposal (Jeremy Howard, Sept 2024), but Ahrefs found 97% of sites with llms.txt received zero bot requests for the file, and Google ignores it. Cheap to add; don’t expect citations from it.
• “Block GPTBot to protect yourself.” Blocking GPTBot stops OpenAI training use — it doesn’t affect citations at all. Live citations come from OAI-SearchBot and ChatGPT-User. Block training bots, allow retrieval bots (prompt 09 checks this).
• “Schema markup boosts AI citations.” Large-scale data shows ~zero measurable impact. Schema is still good practice for search — just don’t expect it to move AI citations.
• “Rank #1 on Google = AI citations.” Only ~30% of ChatGPT citations come from Google’s top-10, and over a quarter of its most-cited pages have zero Google visibility. Rankings help; they’re not the gate.
• Fake-precise statistics. If a guide gives you an exact table (“quotations +41.2%”), be suspicious — secondary sources conflict on the Princeton paper’s per-method numbers. The paper itself reports 30–40% bands. Anyone more precise than the paper is inventing precision.

Use AI for the mechanical 70%. Keep the judgment 30% human — that’s the part clients actually pay for.