How to Get Cited by AI, and What a Citation Is Worth

What it takes to get cited by AI

An AI assistant cites you when it can fetch your page, finds a clear answer on it, and sees other sources naming your business. A citation is a link or source line beside the answer. A mention is your name in the text with no link.

Assistants that show sources first retrieve pages, then write the answer from them. Your job is to be easy to retrieve, easy to quote and easy to verify. The question that matters to you is narrow: on the buyer questions that bring in leads, does an assistant name your business?

Nobody sells a citation, and nobody can promise one. This page gives the numbers, the order of work and the point where you stop.

What a citation is worth in leads

Start with scale. AI referrals were 1.08% of all website traffic across 10 industries, measured over more than 3.3 billion sessions (Conductor, 2026). For most firms this channel is small today, so count leads, not appearances.

Two forces pull against each other. When Google showed an AI summary, people clicked a regular result link in 8% of visits. Without one, they clicked in 15% (Pew Research Center, 68,879 searches, March 2025). Clicks on a link inside the summary itself happened in 1% of visits. Ahrefs measured a 34.5% lower click-through rate for the top-ranking page when an AI Overview appeared (300,000 keywords, 2025).

Being cited helps win some of that back. Brands cited in an AI Overview earned 35% more organic clicks than when not cited, across 3,119 search terms at 42 organizations (Seer Interactive, 2025). Seer says it cannot prove the citation causes the extra clicks. Read a citation as partly defence: it keeps your name in front of buyers who now click less.

One company reports far higher value per visitor. Ahrefs says 0.5% of its traffic and 12.1% of its signups come from AI search, and that those visitors convert 23 times better than organic visitors (2025). Ahrefs warns the result may be unique to it, so do not plan around it.

Here is a worked example. Only the second line comes from a study. Every other input is assumed, so replace it with your own CRM numbers.

• Monthly website visits: 10,000 (assumed).
• Share from AI assistants: 1.08% (Conductor, 2026). That is 108 visits.
• Visit-to-enquiry rate: 3% (assumed). That is about 3 enquiries.
• Close rate 30% and average job $4,000 (both assumed). That is about one job, roughly $3,900 a month.

Where AI answers get their sources

Most citations point to places a business controls. Yext counted 6.8 million citations across three assistants from 1 July to 31 August 2025. 86% came from sources brands already control, such as websites and listings.

For service firms, the Business Profile stands out. Both firms below sell local tools, so read their figures as direction, not law.

What this costs you is hours, not media spend. You own your website, your profile and your review requests. Check this week: ask an assistant three buyer questions from your trade, open the sources it shows, and see whether your business appears on them.

• 28.5%of all citations are Google Business Profile, 1,967,762 citationsBrightLocal, 2026
• About 93%of cited domains are individual business websites, 2026BrightLocal, 2026
• 44%of citations are brand websites, listings 42%, July to August 2025Yext, 2025
• 2%of citations are Reddit and similar platforms once location and query intent applyYext, 2025

Do you need to rank on Google first?

Only in part, and it depends on the assistant. In a dataset of 15,000 prompts, only 12% of links cited by ChatGPT, Gemini and Copilot appear in Google's top 10 results for the same prompt (Ahrefs, 2025). More than 80% of their citations come from pages that do not rank at all for the target query. Perplexity favors pages that rank well, with 28.6% of its cited URLs in the top 10.

Google's own AI Overviews sit closer to its rankings. Ahrefs found 76.10% of cited pages in the top 10, across 1.9 million citations from 1 million AI Overviews (July 2025). Search Engine Journal reported a later study of 863,000 keywords where the figure was 38% (March 2026), and called the two datasets not directly comparable.

Keep your pages indexed and in good order. Do not count on rankings alone to carry you. Check this week: search your best buyer question on Google, then ask an assistant the same question, and compare which sites each one names.

Step zero: let the assistants reach you

A page an assistant cannot fetch cannot be cited. Ask whoever runs your website to check four things this week and give you the answers in writing. It is the cheapest step on this page, and it gates every other step.

1. robots.txt: allow the search crawlers of the assistants your buyers use, plus the two big search engines. Check the file once a quarter, because a site update can block them by accident.
2. Firewall and CDN: security tools often block unknown bots by default. Test a fetch from outside your network.
3. Snippets: Google says a page must be indexed and eligible to be shown in Search with a snippet (Google Search Central, updated December 2025). Remove any nosnippet setting from pages you want quoted.
4. Training versus search: blocking a training crawler is a separate choice from blocking a search crawler. You can do the first and still allow the second (Primal, 2026).

Make every page quotable

Assistants quote passages, not pages. Open each page with the direct answer in the first two sentences. Put the price range, timeline or area served there too. A page titled How much does a kitchen remodel cost in Laval? should open with the range and what drives it, not with the company history.

Make each section stand on its own, so a reader who sees only that block still understands it. Merge thin pages on the same question into one strong page and redirect the weak URLs. Agents that act on your offers also need current prices and policies, as agentic commerce readiness explains.

Add evidence with a source, a year and a sample. Academic researchers tested content changes for generative engines on a lab benchmark and reported visibility gains of up to 40% in generated answers (Aggarwal and colleagues, KDD 2024). Effects vary by topic, so treat it as a reason to test, not a promise.

Publish one original number a year with the method behind it, such as a survey of your own customers. An assistant has something to quote that a general claim cannot offer.

Get named where assistants already look

Ahrefs analysed 75,000 brands and found that brand web mentions had the strongest correlation with AI Overview visibility: 0.664, against 0.218 for backlinks (2025). It is a correlation and covers one feature, so it shows where to look, not what causes a citation.

For a service business, work through this list in order:

1. Complete your Business Profile and match your name, address and phone number on every site and profile.
2. Ask each happy customer for a review on the profile and on the review sites your buyers use.
3. Claim or fix your listings on the directories of your trade.
4. Earn one mention in a trade publication or association, with a named person quoted.

How to get cited in ChatGPT

ChatGPT sent 87.4% of measured AI referral traffic across 10 industries (Conductor, 2026). The share of US ChatGPT prompts that carry citations rose from about 1.6% in June 2025 to roughly 6.8% by May 2026 (Similarweb, 2026).

Take the same steps in the same order. Allow its search crawler, answer first, and get named on the sources it reads. Then ask ChatGPT your buyer questions yourself, several times, and write down what it says.

Plan for visitors to land on your homepage. Roughly six in ten referred visits land there, not on a specific page (Similarweb, 2026). Your homepage and About page must carry the sale.

Which tactics work, and which to skip

Strong means a stated sample and independent sources that agree. Mixed means a sample, but the result is correlation or varies by topic. Weak means no controlled test found, or a test that found no lift. Strongest first. Spend where the grade is strong, test where it is mixed, and stop where it is weak.

Schema and llms.txt: keep, skip or stop

Google says no special schema is needed to appear in AI Overviews or AI Mode, and no additional requirements apply. Semrush found cited pages were more likely to use a few schema types across 5 million URLs, and said this does not prove schema causes citations (2026).

Keep Organization and Person markup to state who you are. Do not pay for it as a citation lever. If a vendor sold you an llms.txt file as a growth tactic, treat the spend as a lesson and move on.

Before you pay anyone for AI visibility, ask in writing for the baseline visibility rate, the list of questions tested and the leads that reached your CRM. A report made of screenshots is not a result.

Why the answer changes every time you ask

SparkToro and Gumshoe ran 2,961 prompts through hundreds of volunteers. The chance of getting the same list of brands twice was below 1 in 100 (reported by Search Engine Journal, January 2026). The study is not peer reviewed, and one of its authors sells tracking.

So one screenshot of your name proves little, and one missing screenshot proves as little. The researchers found that visibility across many queries is more consistent than ranking position.

Measure a rate: the share of answers that name you across a fixed list of questions, run several times. Fifteen questions run five times give 75 answers. Named in 12 of them, your visibility rate is 16%. The method for measuring AI search traffic goes deeper.

How to measure it with tools you already own

You need four views and one monthly report. None needs a new software budget. Report three numbers each month: your visibility rate, your AI-sourced leads and the jobs they became. Compare your AI share of sessions with the 1.08% average (Conductor, 2026).

1. Analytics: in your acquisition reports, filter sessions by the source domains of the assistants your buyers use, and save the view.
2. Search Console: watch impressions against clicks on the pages that answer buyer questions. Falling clicks with steady impressions fit the pattern Pew describes (2025).
3. Bing Webmaster Tools: check whether your account shows an AI performance report (Primal, 2026).
4. CRM: add AI assistant to your lead source list. Ask how people heard of you on forms and first calls. Count leads, booked meetings and jobs with that source.

What to do this week

Five days, one owner each, about nine hours in total. Finish them in order.

1. Day 1, whoever runs your website, 1 hour: confirm search crawlers can fetch your key pages and no nosnippet setting blocks quoting. Get the answer in writing.
2. Day 2, your office manager, 2 hours: complete the Business Profile and match name, address and phone on your site, profiles and top five listings.
3. Day 3, you or your lead writer, 3 hours: rewrite the first two sentences of your three best money pages so they answer the buyer's question with a price range, timeline or area served.
4. Day 4, a team member, 2 hours: write 15 buyer questions, run each 5 times and log the result. Add the CRM lead source and the analytics filter.
5. Day 5, you, 1 hour: pick one original number you can publish within 60 days, name its owner and date, and put the day-90 review in the calendar.

When to stop spending on it

Set the cap before you start. The monthly value of AI-sourced leads in your CRM must beat the monthly cost of the work. In the worked example the channel is worth about $3,900 a month, so spend above that buys a loss.

Review on day 90. If your log shows no assistant names any firm in your category for your buyer questions, and the CRM shows no AI-sourced leads, finish days 1 to 3 and stop. Anyone who asks for more should show you leads in your CRM, not mentions.

Stop paying now for the weak-grade tactics in the cards. Keep paying for crawler access, your Business Profile, reviews and pages that answer first. AI visibility pays back inside a growth system that counts leads and follow-up.