How to get your brand cited by ChatGPT and Perplexity

A six-step system to get into AI answers, built on the two studies that actually measured it, with the guesses labelled as guesses.

Rows of server racks in a data centre
Photo: Esquilo, CC BY-SA 3.0, via Wikimedia Commons

To get cited by ChatGPT and Perplexity, write pages that answer a buyer's question in the first lines and back it with statistics, quotations and named sources, which raised visibility in AI answers by about 30% to 40% in the KDD 2024 GEO study. Then get covered by third-party sites, because AI search cites earned media far more than brand websites: 72% to 92% of GPT's cited sources were earned media in a 2025 study. Measure it by testing the same set of buyer prompts every month and counting how often you are named.

Key takeaways
  • +40.9% AI-answer visibility from adding quotations, +30.6% from statistics (GEO-bench, KDD 2024).
  • Keyword stuffing was the only tested edit that lowered visibility (about -8%).
  • 72% to 92% of GPT's citations in some consumer categories were earned media, not brand sites (Chen et al., 2025).
  • +115% for a fifth-ranked page that added citations: lower-ranked pages gain most.
  • Track share of answers monthly with a fixed prompt set, each prompt run three times.

Generative engine optimisation, or GEO (also called answer engine optimisation, AEO), is the work of making an AI assistant mention and link your brand when it answers a question. It overlaps with SEO but is not the same.

The 2025 study below found that ChatGPT's sources often differ from Google's top results, and that brand-owned pages are cited less in AI answers than in Google. This guide separates what has been measured from what is still a guess. For the mechanics of how engines pick sources, read how AI Overviews and ChatGPT search choose sources.

GoalSEOGEO / AEO
Where you winA ranked list of linksInside a written answer
Unit judgedThe pageThe passage or sentence
Who gets cited mostMixed, brand sites includedThird-party earned media (GPT)
How you measureRank, clicksMentions, citations, share of answers
+40.9%visibility from adding quotations (GEO-bench)
+115%for a 5th-ranked page that added citations
92%earned-media share of GPT citations, US consumer electronics
8% vs 15%Google link clicks with and without an AI summary
The story

After his grandmother died in 2022, Jake Moffatt asked the chatbot on Air Canada's website about bereavement fares. It told him he could buy a full-price Vancouver to Toronto ticket and claim the bereavement discount within 90 days. He paid about CA$1,630, applied, and was refused: the airline's own policy page said the discount could not be claimed after travel. Before British Columbia's Civil Resolution Tribunal, Air Canada argued the chatbot was a separate legal entity responsible for its own actions. Tribunal member Christopher Rivers called that remarkable, held that the airline was responsible for everything on its website, chatbot included, and on 14 February 2024 found negligent misrepresentation, awarding about CA$812 plus interest and fees.

Wikipedia: Moffatt v. Air Canada (2024 BCCRT 149), with links to the tribunal decision

The case was about a company's own bot, but the lesson carries to ChatGPT and Perplexity: an AI answer about your brand is read as fact, and a wrong one is your problem to find. The fix is the same as for citations: clear, consistent, quotable facts on pages the engines trust.

What does the research say gets content cited by AI?

Two controlled studies answer this:

  • the first shows which page edits raise a source's share of an AI answer;
  • the second shows which kinds of sites AI engines prefer to cite.
GEO: Generative Engine OptimizationPranjal Aggarwal, Vishvak Murahari et al. · KDD 2024

The authors built GEO-bench, 10,000 queries across domains, and tested nine page-editing methods on a generative engine based on GPT-3.5, then on Perplexity for 200 queries. Adding quotations raised a source's position-adjusted word share in the answer by about 41%, adding statistics by about 31% and citing sources by about 28%. Keyword stuffing lowered it by about 8%. Gains were largest for pages ranked lower: citing sources raised visibility by 115% for the fifth-ranked source and cut it by 30% for the first. On Perplexity, quotation addition lifted visibility by about 21% to 30%. Limit: engines change fast, and the test engine was a research setup, not ChatGPT itself.

Read the paper (free)
Generative Engine Optimization: How to Dominate AI SearchMahe Chen, Xiaoxuan Wang, Kaiwen Chen, Nick Koudas (University of Toronto) · arXiv · 2025

Using 1,000 consumer ranking prompts in the US and Canada, plus language and paraphrase tests, the authors classified every cited source as brand-owned, social or earned (third-party). Google's mix was balanced, while GPT leaned heavily on earned media: 81.9% of citations for US automotive queries, 77.6% to 92.1% for consumer electronics, and almost zero social content. Overlap between AI and Google sources was often low, for example about 15% of top-5 domains for smartphones. Their advice: invest in earned coverage, structure pages for extraction with tables and schema, and earn coverage in local-language publications. Limit: a preprint, consumer categories only.

Read the paper (free)
Change in AI-answer visibility by page edit (GEO-bench)
See the data as a table
Position-adjusted word count vs baseline (%)
Quotation addition40.9
Statistics addition30.6
Fluency optimisation28.0
Cite sources27.5
Technical terms17.6
Easy-to-understand14.0
Authoritative tone10.4
Unique words6.2
Keyword stuffing-8.3

Source: Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024 (Table 1) · Percent change against the unoptimised baseline of 19.3, calculated from the paper's table.

Step 1: How do I find the prompts my buyers use?

Start with 50 to 100 prompts written the way a buyer would ask an assistant, grouped by stage of the purchase. Chen et al. found engines differ in how sensitive they are to phrasing, so include two or three wordings of each important question.

  1. Category prompts. "Best B2B voice AI platforms for banks in India", "top CRM for a 20-person sales team".
  2. Comparison prompts. "Brand A vs Brand B for collections calls", "alternatives to Brand A".
  3. Problem prompts. "How do I reduce no-shows for counselling calls", with no brand named.
  4. Proof prompts. "Is Brand A reliable", "Brand A pricing", "Brand A reviews".
  5. Local prompts. The same questions in Hindi or another Indian language your buyers use, since AI Mode now answers in Hindi, Bengali, Tamil, Telugu and other Indian languages.

Pull wording from these places:

  • your sales call notes and support tickets;
  • Google Search Console queries of eight words or more;
  • forum threads where buyers ask each other.

Long, question-style searches are where AI summaries appear most: Pew found Google showed a summary for 60% of question-style searches and 53% of searches of ten words or more, against 8% for one- or two-word searches.

Step 2: How do I test where my brand appears today?

Run every prompt in each engine your buyers use, in a logged-out or fresh session, and record five things in a sheet:

  1. whether you are mentioned;
  2. whether you are linked;
  3. your position in the answer;
  4. which competitors are named;
  5. which URLs are cited.

Repeat each prompt three times, because answers vary run to run. The cited URLs are the real output: they tell you which publications, directories and pages the engine trusts for your category.

Read what the answer says, not just whether you appear

Log the claims an engine makes about you (price, policy, features), because buyers act on them, as the Air Canada case above shows.

A large newspaper printing press at the Western Daily Press in 1908
AI engines lean on the press again. Third-party coverage is the citation source they trust most.Photo: Samuel Loxton (artist), Public domain, via Wikimedia Commons

Step 3: How should I rewrite my pages for AI answers?

Rewrite your key pages so that each section opens with a quotable, specific answer and carries numbers, named sources and a short attributed quote. These are the three edits with the largest measured effect in the GEO study. Concretely:

  • Answer first. Put a two- or three-sentence answer at the top of the page and at the top of every section, with the main noun phrase in it.
  • Add statistics. Replace "fast" with the number, its unit and its date. Engines extract sentences; a sentence with a figure is more extractable.
  • Add quotations and citations. Quote a customer, an analyst or a standard, and link the source. Citing sources helped low-ranked pages most.
  • Use tables. Chen et al. recommend comparison tables, pros and cons lists and explicit specifications, plus Schema.org markup for prices, specs, availability and reviews.
  • Make the entity clear. State who you are, what you sell, where, and for whom in plain words on the home, about and product pages, using the same name everywhere.
  • Skip keyword stuffing. It was the only tested method that reduced visibility.

Step 4: How do I get mentioned on the sites AI engines cite?

Earn coverage on the third-party pages that the engines already cite for your prompts, because AI search draws most of its citations from earned media. Use the cited-URL list from Step 2 as your outreach list:

  1. Trade and business press. Give journalists data they cannot get elsewhere: a benchmark from your own anonymised usage, a survey, a teardown. Data gets quoted.
  2. Comparison and review sites. Make sure your listing on the directories and review sites cited in your category is complete and current, with real customer reviews.
  3. Expert lists and guides. Pitch a useful contribution to the "best X for Y" articles that appear in citations, rather than asking for a mention.
  4. Local-language publications. Chen et al. found translating your own site is not enough for GPT-like engines; coverage in local-language publications matters.
  5. Partners and customers. Joint case studies published on a customer's or partner's site count as earned media.

The domains that recur across AI answers are ranked in the top domains AI search engines cite.

Share of answersmeasured monthly1. Prompt list2. Test engines3. Fix pages4. Earn mentions5. Local language6. Track and repeat
The loop: every monthly test tells you which pages to fix and which sites to earn coverage on next.

Step 5: Which tactics are proven and which are guesses?

Only a handful of tactics have been tested in controlled studies; most advice about AI search is extrapolation. Here is the honest split.

TacticEvidenceStatus
Add statistics, quotations, cited sources+28% to +41% visibility, GEO-bench (KDD 2024)Tested
Keyword stuffing-8% visibility, GEO-benchTested, harmful
Earned media coverage72% to 92% of GPT citations are earned media (Chen et al. 2025)Measured correlation
Comparison tables and Schema.org markupRecommended by Chen et al. from source analysisPlausible, not isolated
Local-language coverageChen et al. cross-language testsMeasured for GPT-like engines
Ranking on Google firstLow overlap between AI and Google sources in many categories (Chen et al.)Helps, not sufficient
Special files or tags "for AI"No controlled study foundGuess
Posting volume on social platformsGPT cited almost no social content (Chen et al.)Weak for citations

Step 6: How do I track share of answers every month?

Re-run the same prompt set on the same day each month and compute three rates:

  • mention rate: how often you are named;
  • citation rate: how often you are cited with a link;
  • share of answers: your share of all brand mentions.

Keep the prompts fixed so changes reflect your work, not a new question list.

Share of answers: a worked month150 prompt runs (50 prompts x 3)30 runs name you: 20% mention rate12 link you: 8%All brands were named 400 times; 30 were youShare of answers = 30 / 400 = 7.5%
The calculator below runs the same arithmetic on your own counts. Fixed prompts, three runs each, same day every month.

Your AI share of answers

Also watch the traffic side. Pew's analysis of 68,879 Google searches by 900 US adults found people clicked a traditional result on 8% of visits when an AI summary appeared, against 15% without one, and clicked a link inside the summary only 1% of the time. Being cited matters more as clicks fall, so measure brand mentions and branded search alongside referral visits.

Server racks in the Indian Railways 139 helpline server room
Answers are now assembled by machines from many sources at once. Being one of those sources is the new shelf position.Photo: Indrajit Das, CC BY-SA 3.0, via Wikimedia Commons

How long before results show?

Page edits can show up quickly in engines that fetch live web pages at answer time, while earned coverage builds over months. Plan a 90-day cycle:

MonthWork
1Prompt list, baseline test, page rewrites
2Outreach to the cited sites
3Re-test and compare

For a reality check on what brands have actually achieved, see does AEO/GEO work. B2B buyers already shortlist vendors before contacting them, increasingly with AI help, so this matters most for companies selling to businesses; see B2B buyers decide before they call you.

My view is that AI should cut repetitive production so a small team can spend its time on work like this; how I use it is on applied AI, and here is how to work with me.

Questions people ask

Has anyone actually got results with AEO or GEO?

In controlled tests, yes: adding quotations, statistics and citations raised visibility in AI answers by about 28% to 41% on GEO-bench, and by up to 115% for lower-ranked pages. Real-world results depend on earned coverage too, which takes longer than page edits.

How do I optimise a page to rank in AI search results?

Answer the question in the first lines, support it with numbers, a short attributed quote and links to sources, add a comparison table where buyers compare options, and keep your brand name and description consistent across your site.

Does ranking on Google mean ChatGPT will cite me?

Not necessarily. Chen et al. found low overlap between the domains GPT cited and Google's top results in several categories, and a much stronger AI preference for third-party sources. Good SEO helps, but earned media carries more weight in AI answers.

Can I pay to appear in ChatGPT or Perplexity answers?

Not in organic answers. Separately, Google began showing labelled sponsored ads inside AI Mode in India in October 2026, so paid placement now exists in some AI search products, clearly marked as ads.

Sources

  1. Aggarwal et al.: GEO: Generative Engine Optimization (KDD 2024) (2024)
  2. Chen, Wang, Chen and Koudas: Generative Engine Optimization: How to Dominate AI Search (2025)
  3. Pew Research Center: Google users are less likely to click on links when an AI summary appears (2025)
  4. TechCrunch: Google's Search Live comes to India, AI Mode gets more languages (2025)
  5. Beebom: Google AI Mode is getting ads in India (2026)
  6. Wikipedia: Moffatt v. Air Canada, 2024 BCCRT 149 (accessed Nov 2026; source for the chatbot story)
AEOGEOChatGPTPerplexityAI searchcontent strategy
Sapta

Sapta (Saptarshi Sarkar) is a mechanical engineer turned marketer in Bengaluru and a solo touring rider. He writes the seven Saptaverse blogs. Work with him or read more of Marketblog.

Keep reading in Marketblog

From the other blogs