How to get cited by AI, and how to know it worked
The one controlled study of this found that quotes, statistics and named sources lifted AI visibility by 30–40% — and that the researchers got there by telling a model to invent them. Not markup. Not a file in your root directory. This is the practical article: what the evidence supports, what it costs, how long it takes, and how to check whether any of it moved — including the Search Console report Google shipped in June that most people don't know exists.
Do local SEO properly and write accurately. Whitespark's 2026 conclusion is that local search signals and AI search signals have effectively merged — the same inputs drive Google, Google Maps, ChatGPT and Perplexity. The two things with the strongest direct evidence behind them are citing real statistics from named sources and keeping pages updated. Both are free. Measurement is the genuinely hard part, and the honest method is prompt testing plus asking new customers how they found you.
What the one controlled study actually measured
A study from Princeton, Georgia Tech and IIT Delhi, presented at KDD 2024, tested generative-engine optimization methods directly rather than looking for correlations. It remains the most useful piece of research in this field.[1]
| Method tested | Visibility change |
|---|---|
| Adding quotations from credible sources | +41% — the strongest single method[1] |
| Adding statistics | +31%[1] |
| Citing sources | +28%[1] |
| Those three together, as the authors summarize them | 30–40% on the paper's main visibility metric[1] |
| Keyword stuffing | Little to no improvement — statistically indistinguishable from doing nothing[1] |
So the pattern is clear enough: content that carries quotes, numbers and named sources gets cited more than content that doesn't. No tooling, no subscription, no specialist.
And keyword stuffing is worth a line of its own. It didn't help. Across five runs it scored statistically level with unoptimized content.[1] Google's own guidance says the same, warning against content made primarily for search traffic and against extensive automation across many topics.[2]
⚠ But read how they tested it, because almost nobody does
The researchers produced those gains by prompting a language model to modify web pages — and the prompts told it to make things up.
The statistics prompt, printed in the paper's appendix, says: "Addition of fake data is expected… Add positive, compelling statistics (even highly hypothetical)." The citation prompt says: "You may invent these sources but ensure they sound plausible." The quotation prompt says to add quotes "even though fake and artificial."[1]
That changes what the study proves. It did not measure whether accurate, checkable numbers earn more citations. It measured whether content shaped like it has evidence gets cited more — and the answer was yes, whether or not the evidence was real.
Which is the actual argument for doing it honestly. If fabricated statistics work about as well as real ones on the machine, then the only thing separating the two is whether a human can check them. Every article on this site prints its sources at the bottom — this one has fifteen — and that was a decision about honesty, not a ranking tactic. It happens to produce the same shape the study rewarded. It just also survives being checked.
I'd rather tell you that than claim a paper validated something it never tested.
Content decays, and there's a number on it
This is the second-strongest finding, and it's the honest argument for maintenance rather than a one-time project.
| Finding | Source |
|---|---|
| AI-cited content runs about 25.7% fresher than organic top-10 results, across nearly 17 million citations | Freshness analysis, 2026[3] |
| Pages not updated quarterly are 3x more likely to lose AI citations over time | AirOps, 2026[4] |
| Content not refreshed in over six months is at a growing disadvantage in citation pools | BrightLocal, 2026[5] |
That's a quarterly job, not a monthly one, and for most local businesses it's a short list: check the service pages still describe what you do, update anything with a year in it, replace numbers that have moved.
One more structural finding worth a mention: 68.7% of ChatGPT-cited pages follow logical heading hierarchies.[4] Not a trick — just headings that describe the section under them, in order. Any competent build does this by default.
The answer nobody in this industry wants to give
Whitespark has run the longest-standing local search ranking factors study in the field. Their 2026 conclusion is short and expensive to say out loud.
Local search signals and AI search signals have effectively merged. The same inputs drive visibility across Google, Google Maps, ChatGPT and Perplexity.[6]
Google's own May 2026 guidance says the same thing in different words: AI Overviews and AI Mode are "rooted in our core Search ranking and quality systems", and to the question of whether SEO still matters for AI search, "in short, yes!"[2]
So here is the practical list, and it should look familiar because it's just local SEO.
Finish the Google Business Profile
Correct primary category, every service written out, service areas, hours, real photos, and reviews coming in steadily. It is still the largest single factor group in local search, it decides where you sit on Google Maps, and it costs nothing. The checklist is here, no email required.
Be listed accurately in the directories that already rank
Directories carry 46.3% of ChatGPT's citations on "best of" queries, which are the ones people ask when they're choosing. Same name, same address format, same phone, everywhere.
Write things worth citing, with the numbers attached
The 41% finding in practice. If you publish anything at all, put real figures in it and name where they came from. A contractor writing honestly about what a roof replacement costs in their county, with the source, is doing the highest-evidence thing available.
Refresh it quarterly
Twenty minutes, four times a year. Prices, dates, anything that's moved.
Exist where your customers talk
Honestly and under your own name. Manipulating AI answers has been spam by Google policy since May 2026,[7] so no bought mentions and no fake endorsement.
If you serve Spanish-speaking customers, read this part twice
This is the finding I'd have most wanted to know a year ago, and almost nobody in local search is writing about it.
Peec AI analyzed over 10 million ChatGPT searches to see how the model finds its sources. When someone asks a question in Spanish, ChatGPT runs its own background research to answer — and nearly half of that research happens in English anyway.[16]
The consequence is blunt: a non-English prompt gets split into English queries before anything is retrieved.[17] So a business that exists only in one language can be invisible to a customer asking in the other — and local businesses end up losing to national ones in their own market.[16]
And there's a number on what bilingual content is worth
A study analyzing 1.3 million AI citations compared translated and untranslated sites directly.[18]
| Citations, Spanish queries | Citations, English queries | Gap | |
|---|---|---|---|
| Untranslated Spanish sites | 17,094 | 2,810 | 431% |
| Translated sites (Spain) | 10,046 | 8,048 | 22% |
Translating closed a 431% gap down to 22%. Visibility in English searches rose by up to 327% compared with untranslated sites, and for translated sites ChatGPT showed virtually no language bias at all — Spanish sites actually received marginally more citations in English than in Spanish.[18] The researchers' summary was three words: untranslated means invisible.
One trap worth naming
Machine translation isn't enough, and there's evidence for that too. Tested on regionally variable vocabulary — "straw" is pajilla, popote, pitillo or bombilla depending on the country — ChatGPT consistently defaulted to the most globally common version, usually Mexican Spanish, and couldn't be reliably steered even with explicit context.[19] A study across nine language models and seven Spanish varieties found most collapsed them into a generic register.[19]
Which matters here more than it sounds. Fellsmere is 77.9% Hispanic, and the Spanish spoken across the Treasure Coast isn't one dialect. Copy that reads as translated-by-software reads as translated-by-software to the person you're trying to reach, whatever the AI thinks.
⚠ The honest limit on this evidence. These studies looked at Spain, Mexico and European-language markets — not US Hispanic markets specifically. The mechanism is the same and the direction is consistent across every platform tested, but I haven't found a study measuring this for a bilingual town in Florida, and I'm not going to imply one exists. What I can say is that the underlying behavior — English-first background retrieval — doesn't change at a state line.
For a business serving both languages here, the practical version is short: have real pages in both, written by someone who speaks the language rather than a plugin, and link them to each other properly. Toda la información en español está aquí — Spanish is my first language, not a course I took, and it costs nothing extra.
How to actually measure it
This is the part almost every guide skips, because the honest answer is that it's awkward. Traditional rank trackers are blind to AI visibility, and analytics under-report AI referrals badly.
The one Google shipped in June
On June 3, 2026, Google announced Search Generative AI performance reports in Search Console — but it's rolling out to a subset of sites, so you may not have it yet.[8] Go and look; it's free and it's the only first-party data that exists. If it isn't there, watch ordinary Search Console performance, conversions, and whether people are searching your business name more often.[8]
Prompt testing, which anyone can do
Write down 10 to 20 prompts a real customer would type — "best plumber in Sebastian FL", "who does emergency AC repair near Vero Beach", "affordable web designer Treasure Coast". Run them across ChatGPT, Perplexity and Google's AI Mode. Note whether you appear, and who does instead.
Repeat monthly, same prompts, same list. That's it. Two points worth knowing before you start:
- Run them signed out. Your own history skews results, which is the same mistake people make checking their Google rankings.
- Don't average the engines. Only 11% of domains cited by ChatGPT are also cited by Perplexity. A single "AI visibility score" is combining systems that disagree 89% of the time.
The question that beats all of it
Ask every new customer how they found you, and write the answer down. It costs nothing, it works when analytics don't, and it's the only measurement that connects to money. If "I asked ChatGPT" starts appearing in that list, you'll know before any dashboard tells you.
What to expect, and when
Nobody can promise you a citation, and anyone who does is guessing. What can be said honestly is roughly how long each part takes to move.
| What | When it typically moves |
|---|---|
| Google Business Profile services added | Ranking shifts observed within 24–72 hours[9] |
| A wrong primary category corrected | Days |
| Directory and listing consistency | Weeks |
| Review velocity showing up in rankings | 2–3 months |
| Local search generally | 3–6 months, compounding through the first year |
| AI citations | Unknown — nobody has published a reliable timeline, and I'm not going to invent one |
That last row is the honest one. This channel is roughly two years old, the engines change their retrieval behavior without notice, and consumer trust in AI search fell 28 points in a single year. Anyone quoting you a ninety-day AI visibility guarantee is selling certainty that doesn't exist yet.
What not to buy
Pulling the series together, here's the list of things sold as AI-search necessities that the evidence doesn't support — with what Google itself says next to each.
| Sold as | The evidence |
|---|---|
| Special AI schema markup | Google: "no special schema.org markup you need to add."[2] A controlled test of 1,885 pages found no citation change.[10] Both sides, in detail. |
| llms.txt | Scored 2.0 out of 10 in a 54-study meta-analysis.[11] Google gives it no special treatment.[2] |
| Content chunking for parsers | Google: "no requirement to break your content into tiny pieces" and "no ideal page length."[2] |
| Press releases for AI visibility | Across 3,600 prompts in 10 industries, a negligible share of AI news citations.[12] |
| A single "AI visibility score" | The engines overlap on only 11% of cited domains.[13] One number hides more than it shows. |
| $3,000–$6,000/month retainers built on the above | Tooling for this reportedly starts at around €49/month.[14] |
And the fair version, because I'm inside this industry too. AI search is real — AI Overviews appear on roughly 40% of local business queries and about 45% of consumers use AI tools for local recommendations.[15] Preparing for it is sensible. The objection isn't to the channel; it's to being charged specialist prices for work that is either free, already part of local SEO, or unsupported by evidence. I corrected my own plan wording over this, so it isn't a rule I'm applying only to other people.
Questions people ask about this
How do I get my local business cited by AI?
The evidence points to four things, none of which require special tooling. Controlled testing presented at KDD 2024 found that adding quotations, statistics and cited sources lifted visibility by 30 to 40%, with quotations strongest at 41% and statistics at 31%. Note the method: the researchers produced those gains by prompting a model to add data, including invented data, so the study shows that evidence-shaped content gets cited more — not that accurate evidence does. Keeping pages updated matters: pages not refreshed quarterly are three times more likely to lose citations. Being listed accurately in directories matters because they carry 46.3% of ChatGPT citations on subjective queries. And a complete Google Business Profile remains the largest single factor group in local search.
How do I check whether AI mentions my business?
Write down 10 to 20 prompts a real customer would type, run them across ChatGPT, Perplexity and Google's AI Mode while signed out, and note whether you appear and who appears instead. Repeat monthly with the same prompts. Google also announced Search Generative AI performance reports in Search Console on June 3, 2026, though these are rolling out to a subset of sites. The most reliable measurement remains asking every new customer how they found you and writing it down.
How long does it take to show up in AI search?
Nobody has published a reliable timeline and anyone quoting one is guessing. Related timings are better understood: Google Business Profile service listings have produced ranking shifts within 24 to 72 hours, review velocity shows up over two to three months, and local search generally moves over three to six months and compounds through the first year.
Does having a Spanish version of my website help with AI search?
The evidence says yes, substantially. A study of 1.3 million AI citations found untranslated Spanish sites received 17,094 citations for Spanish queries but only 2,810 for the same queries in English, a 431% gap. Translated sites narrowed that gap to 22%, with visibility in English searches rising by up to 327%. Separately, an analysis of over 10 million ChatGPT searches found that when a question is asked in Spanish, nearly half of the model's background research is still conducted in English. Machine translation is not sufficient, as models default to a generic or Mexican Spanish register rather than local usage.
Is a single AI visibility score worth tracking?
Not on its own. An analysis of 680 million citations found only 11% of domains cited by ChatGPT are also cited by Perplexity, so a score that averages the engines is combining systems that disagree with each other most of the time. Tracking each engine separately against a fixed list of prompts is more informative.
Do I need to publish content to be cited by AI?
Not necessarily. Directories, review platforms, community sites and video carry the majority of citations, and 82% come from earned media rather than a brand's own pages. Publishing helps if what you publish is accurate and cites its sources, since that was the strongest measured lever. Publishing large volumes of thin content does not help and may hurt, as keyword stuffing was measured as reducing visibility.
- Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande, "GEO: Generative Engine Optimization," KDD 2024 (arXiv:2311.09735v2) — GEO-bench, 10,000 queries, nine methods, five random seeds. Table 1, Position-Adjusted Word Count against a 19.3 baseline: Quotation Addition 27.2 (+41%), Statistics Addition 25.2 (+31%), Cite Sources 24.6 (+28%); the authors summarize these three as a 30–40% relative improvement. Keyword Stuffing scored 19.8 against a 19.8 baseline across five seeds. ⚠ The optimization prompts printed in Appendix B.4 instruct the model to add fabricated statistics, invented sources and artificial quotes.
- Google, Google's Guide to Optimizing for Generative AI Features on Google Search, Search Central, May 15, 2026 (authored by John Mueller) — including the mythbusting section on structured data, llms.txt and content chunking, and Google's people-first guidance on content made primarily for search traffic.
- Content freshness analysis across approximately 17 million AI citations, 2026 — relative freshness of cited content versus organic top-10 results.
- AirOps, 2026 State of AI Search — citation decay for pages not updated quarterly, and heading-hierarchy share among ChatGPT-cited pages.
- BrightLocal, 2026 Local Search Industry Report — disadvantage in AI citation pools for content not updated in over six months.
- Whitespark, Local Search Ranking Factors, 2026 — conclusion that local search signals and AI search signals have effectively merged.
- Google Search Central, May 2026 — spam policy position on manipulating AI answers.
- Google, announcement of Search Generative AI performance reports in Search Console, June 3, 2026 — rolling out to a subset of websites; guidance on alternative measures where unavailable.
- Sterling Sky — observed local ranking shifts within 24 to 72 hours of adding Google Business Profile service listings.
- Ahrefs, 2026 — 1,885 pages tracked after adding JSON-LD between August 2025 and March 2026, showing no measurable citation change in the following 30 days.
- Cyrus Shepard, Zyppy, May 2026 — meta-analysis of 54 experiments, patents and case studies scoring 23 AI citation factors, including llms.txt at 2.0 out of 10.
- BuzzStream and Citation Labs, 2026 — analysis across 3,600 prompts and 10 industries finding press releases captured a negligible fraction of AI news citations.
- Citation analysis across ChatGPT, Google AI Overviews and Perplexity, 2026 — 680 million citations examined for cross-platform domain overlap.
- LLM Pulse, GEO Agency Guide, August 2026 — stated agency tooling costs for delivering AI search optimization services.
- Sterling Sky, 2026 — share of local business queries returning AI Overviews, and consumer use of AI tools for local recommendations.
- Peec AI, 2026 — analysis of over 10 million ChatGPT searches examining the language in which the model conducts background research when answering non-English prompts.
- Search Engine Land, 2026 — analysis of multilingual AI visibility, including the splitting of non-English prompts into English fan-out queries prior to retrieval.
- Weglot, 2025–2026 — study of 1.3 million AI citations comparing translated and untranslated sites. ⚠ The 431% figure is specific to GOOGLE AI OVERVIEWS; ChatGPT showed a far milder gap of about 3.5%. The Spain sample was 98 untranslated sites and 83 translated. ⚠ Weglot sells website translation software, and the Search Engine Journal write-up of this study carries a paid-sponsorship disclosure across AI Overviews, AI Mode, ChatGPT and Perplexity in Spanish-language markets.
- Search Engine Land, March 2026, "the Global Spanish problem" — testing of regionally variable Spanish vocabulary across GPT models, and an evaluation of nine language models across seven Spanish varieties.
Most of this list is free. Let me tell you which part you're missing.
Send me your business name and I'll run the prompts, check the directories and look at your Google listing. You'll get the list either way, and if the honest answer is that you don't need to pay anyone for this yet, that's what I'll tell you.
