Skip to content

An agency will charge only when AI cites you. Is pay-per-citation GEO pricing a good deal?

Most AI answers link to nothing, so a per-citation fee can pay for easy prompts. The clauses that make pay-per-citation GEO pricing fair, and when to avoid it.

By Published 15 min read
Costs and buying, 15 min read — A long dark counter receding into fog, with only a few small glowing blue-violet tokens resting on it far apart.

The short answer

Pay-per-citation GEO pricing is rarely a good deal as first offered. Most AI answers carry no link, so a fee per citation rewards the cheapest prompts rather than your buyers' questions. Accept one only with a fixed prompt set, branded prompts excluded, a repeatable count, a baseline and a cap. Otherwise prefer a fixed fee with part at risk.

Key takeaways

  • In Similarweb's US data, about 6.8% of ChatGPT answers carried a citation in May 2026, so a fee charged per citation pays out on a small minority of the answers buyers read.
  • Ahrefs found that 14.4% of 66.6 million free, logged-out ChatGPT responses in August and September 2026 triggered a web search, and an answer that does not search has nothing fresh to cite.
  • A per-citation contract is only as fair as its prompt list: branded prompts and questions nobody types are cheap citations that cost the agency little and bring you little.
  • Agree the engines, sessions, country, runs, counting unit and raw evidence before the first invoice, and pay only for citations above a baseline month counted the same way.
  • A fixed fee with a defined share at risk against a measured share of answer usually aligns the agency with your buyers better than a price per link.

What does a pay-per-citation offer actually charge for?

It charges for a link, not for a recommendation. In a pay-per-citation deal, the agency invoices a fixed amount each time an AI answer links to or attributes your site, usually with a small monthly fee or none. The pitch is easy to like: you pay only when it works. Whether it works depends on what the contract counts, and the pitch rarely says.

Three versions turn up in proposals. Pay per citation counts answers that link to your domain. Pay per mention counts answers that name your brand, linked or not. Pay per placement counts appearances on a named list of prompts. Each pays for a different event, and the effort needed to produce each one varies a great deal, so read the definition before you read the price.

Citation
An AI answer that links to or attributes a source on your site. A mention names your brand in the answer, with or without a link.

Prior work and a disclosure

Sigzen AI sells AI visibility programmes at published prices, and part of our fee is at risk against a measured result, so we hold a view on how fees should be tied to outcomes.11 Read what follows with that interest in mind. The checks below work whoever you buy from, and most of them fit in a contract annex.

Others have written about how this market prices its work. WebFX, which sells GEO services, published a guide to what GEO costs, and Tenten published a benchmark of GEO pricing models.12 13 Our own survey of published prices covered agency retainers and tool subscriptions.6 What this post adds is a clause-by-clause reading of the per-citation offer itself.

Why a citation is a thin thing to buy

Most AI answers link to nothing, so a citation is a rare event, and a price per rare event pushes the work towards whatever produces it most cheaply. Similarweb, which sells web analytics, found that about 6.8% of US ChatGPT answers carried a citation in May 2026, up from about 1.6% in June 2025.1 The rest named brands, or none, without a link.

The spread between sectors is wide. In the same May 2026 data, about 23% of travel and hospitality answers carried a citation, against under 4% in professional services.1 An accounting firm on a per-citation deal is buying from a shelf that is nearly empty; a hotel group is buying from a full one. The same price per link means very different things in the two markets.

Search is the gate. Ahrefs, which sells SEO and AI visibility tools, studied 66.6 million free, logged-out ChatGPT responses from August and September 2026, across all countries, and 14.4% triggered a web search.2 An answer written from the model's memory has nothing fresh to cite. Per-citation fees only touch the searched minority of answers.

Engines differ as well. Semrush, which sells an AI visibility toolkit, analysed 126 million US prompts from January to April 2026 and found that ChatGPT cites about 15 sources per response, against about 3 for Gemini.3 Fifteen slots are easier to land in than three. A per-citation count therefore rewards work on the engine with the most slots, whichever engine your buyers actually use.

How often AI answers cite at all

Citations are the exception, not the rule

6.8%of US ChatGPT answers carried a citation in May 2026Similarweb, Jun 2025–May 2026 1
14.4%of free, logged-out ChatGPT responses triggered a web searchAhrefs, Aug–Sep 2026 2
~15 vs ~3sources per response, ChatGPT vs Gemini, US promptsSemrush, Jan–Apr 2026 3
Sources: Similarweb, AI search stats 2026; Ahrefs, ChatGPT query fan-out study; Semrush, 2026 AI Visibility Index. All three companies sell analytics or visibility tools.

Which prompts would the agency choose to work on?

The ones that cost least to win, unless the contract says otherwise. A per-citation price is a price per event, and the cheapest events are prompts that contain your brand name, prompts that nobody types, and prompts on engines that cite generously. None of those is where a buyer builds a shortlist of suppliers.

Branded prompts are cheap citations

A question that names your company points the engine straight at you. Profound, which sells AI visibility tools, analysed 11.84 billion citations across eight engines from 16 April to 16 July 2026 and found that about 57% went to company-owned sites.4 When the prompt contains your name, your own site is the obvious source to cite, and the agency did little to earn the link.

Branded searches are also where AI summaries now appear by default. In Ahrefs' US desktop data, 73.3% of branded keywords triggered a Google AI Overview in September 2026, as a monthly average.5 A buyer who types your name already knows you. A citation in that answer shows the engine found your site; it does not show that anyone new put you on a list.

Questions nobody asks

The second cheap source is the obscure prompt. A question phrased exactly the way your product page is phrased will often cite that page. It counts as a citation and reaches nobody. The prompts worth paying for are commercial ones: in Profound's study of 7.5 million en-US ChatGPT conversations from 23 June 2025 to 22 June 2026, the share with commercial intent rose from 13.9% to 19.2%.9 Those prompts are where rivals compete, and they are the hardest to win.

Our view: the prompt list is the real price. Two deals at the same price per citation, one counted on branded prompts and one on unbranded buyer questions, differ in value by far more than any discount an agency can offer on the rate.

Where per-citation effort tends to go

Value to you →

Category prompts you already leadWorth keeping, but you would be paying again for links you have.
Unbranded buyer questionsHard to win and where shortlists form; the prompts worth paying for.
Branded and obscure promptsCheap citations with little new reach; exclude them or price them at zero.
Prompts on engines your buyers skipEffort spent where few of your buyers look.

Cost to win a citation →

Illustrative. A per-citation price pulls work to the left-hand column unless the contract fixes the prompt list; the quadrant to pay for is top right.

Who counts the citations, and how?

Whoever counts controls the invoice, so the count has to be one you can repeat yourself. AI answers change between runs of the same prompt. A citation that appears on Tuesday can be gone on Wednesday with nothing changed on your site, and an agency that counts on its best day will always find more than you do.

Repeat runs help, though less than they seem. MaxAEO, which sells AI visibility measurement, published guidance treating repeat runs of one prompt as clustered: they resemble each other, so each extra repeat adds less information than a new prompt would.7 A count across a fixed set of prompts, each run on separate days, is fairer than many runs of a handful of prompts.

Write the conditions down. If the agency's tool is the only record, you cannot dispute an invoice, because nobody else saw the answers. Ask for the answer text and the cited URL for every counted citation, with the time it was captured, and keep a copy on your side.

TermWhat the contract should say
EnginesA named list, matched to the engines your buyers use
SessionsSigned out, clean, no chat history or memory
Country and languageFixed per prompt and matched to your buyers
RunsEach prompt on three separate days per billing period
Counting unitOne citation per prompt per period, not per run
EvidenceAnswer text, cited URL and timestamp, shared monthly

Illustrative. Terms the Sigzen AI editorial team would ask for in any per-citation contract; adjust the engines and countries to your market.

Agree what one citation is

If a prompt runs three times in a month and cites you each time, is that one citation or three? Counted per run, the agency earns three times for the same work, and adding runs adds revenue. Counted once per prompt per period, the unit maps onto a share of answer, which is the number you want to move in any case.

Our view: if an agency will not agree a counting unit in writing before work starts, the offer is not priced per citation. It is priced per whatever the agency's tool happens to report that month.

What is a citation worth to your business?

Less than the click it might bring, and more than nothing. Clicks from AI answers are still a small channel for most sites. Work out what a cited answer is worth in your own funnel before you agree a price per citation, because the agency's rate is set by its costs and margins, not by your revenue.

The click volumes are modest. In SE Ranking's study of 101,574 websites in 250 countries, AI search sent 0.32% of all visits in January–April 2026.14 Being cited does lift clicks where it happens. In Seer Interactive's panel of 53 brands from January 2025 to February 2026, pages cited in an AI Overview earned about 120% more organic clicks per impression than when they were not cited.8

Influence counts for more than clicks with most B2B buyers. In G2's survey of 1,076 B2B decision makers in North America, EMEA and APAC (March 2026), 51% said they start software research in an AI chatbot more often than in Google.10 Much of that influence comes from being named, linked or not. A per-citation contract ignores the mention that put you on a shortlist and pays for a link that may never be clicked.

Our view: if you must price a citation, price it from your own numbers. Take the visits a cited answer sends, your visit-to-lead rate and the value of a lead, and keep the rate well below what that chain is worth. Our business case guide walks through that chain step by step, and our note on mentions and citations covers which of the two to report.

A long dark counter with a few small glowing tokens scattered along it, most of the surface empty, lit by a single cool overhead light.
A price per link pays on the few answers that carry one.

How does pay-per-citation compare with a retainer?

It moves the risk rather than removing it. A fixed retainer risks paying for work that does not move the number. Pay per citation risks paying for numbers that do not matter. A fixed fee with a defined share at risk sits between the two: the agency is paid for the work, and earns the rest only if a measured result moves.

Fixed prices are what agencies publish. Across the 13 listed agencies pricing in US dollars that we recorded on 5 and 6 October 2026, the median monthly entry price was $3,400, and listed tools had a median entry tier of $79.6 If a per-citation rate is not on the agency's pricing page, compare it with the retainer prices that are, at the volume of citations you could realistically expect.

Three ways to tie a GEO fee to results

Pay per citation

  • Little or no fixed fee
  • Paid per link counted
  • Pulls work to cheap prompts
  • Invoice moves when engines change

Fixed fee, share at risk

  • Work paid for, a defined share at risk
  • Target set on a fixed prompt set
  • Measured like the baseline
  • Rewards unbranded buyer prompts

Fixed retainer

  • Same fee every month
  • Deliverables, not results
  • Simple to budget
  • No direct link to the number
Illustrative. The highlighted column is the structure the Sigzen AI editorial team recommends for most buyers; each can work with a tight enough contract.

Our programme is one version of the middle column. On Growth and Scale, 20% of the monthly fee is at risk against the share-of-answer target set at the audit, and is invoiced only for a quarter in which the target is met, as on our pricing page.11 Scale also carries a 90-day citation guarantee with published qualifying criteria. Other designs exist; what matters is that the share at risk is defined, and measured the same way as the baseline.

Clauses to write in before you sign

Six clauses turn a per-citation offer into something you can audit. An agency confident in its work will usually accept them. Resistance to any one of them tells you something about how the count was going to be made.

Six clauses for a per-citation contract

  1. Fixed prompt set

    Unbranded buyer questions agreed before work starts and frozen for the billing period.

  2. Branded prompts out

    Prompts containing your name are excluded from the count or priced at zero.

  3. Measurement protocol

    Named engines, clean signed-out sessions, fixed country and language, separate days.

  4. Counting unit

    One citation per prompt per period, never one per run.

  5. Baseline and evidence

    A counted month before fees start; answer text, URL and timestamp for every citation billed.

  6. Cap and exit

    A monthly ceiling on fees and a short notice period.

Illustrative. Clause list from the Sigzen AI editorial team; have your own lawyer word them for your contract.

The baseline clause saves the most money. Count citations under the agreed protocol for a month before any work starts, and pay only for citations above that count. Without it, you pay for links you already had, including the ones your own team earned last year.

The cap matters because engines change without notice. OpenAI moved paid ChatGPT plans to GPT-6 Sol from 7 October 2026 and Free and Go users to GPT-6 Luna from 8 October.15 A change like that can shift how often answers cite for every brand at once. Without a ceiling, your invoice moves with it, and no work was done to earn the difference.

Questions to ask in the pitch meeting

  • Which prompts will you count, and who wrote them?
  • Do prompts that contain our brand name count?
  • Is a citation counted once per prompt or once per run?
  • Can we see the raw answers for every citation you bill?
  • What happens to the fee if an engine starts citing more sources?

A good answer to each is short and already written in the draft contract. Our RFP question list covers the wider vetting of an AI visibility provider, and our measurement spec sets out what any vendor's count should document.

When is pay-per-citation a reasonable deal?

When the citation is the outcome you actually want and the count is clean. That is more likely in sectors where answers often link, such as travel and hospitality, where about 23% of US ChatGPT answers carried a citation in Similarweb's May 2026 data.1 There, a cited answer is a plausible route to a booking, and a price per citation tracks something real.

It also works as a small top-up on a fixed fee: a bonus for each new unbranded citation above baseline, capped, and counted under the agreed protocol. That keeps the agency paid for the work while giving it a reason to aim at the harder prompts.

It is a poor deal where answers rarely link, such as professional services, or where your buyers decide on names rather than links. Our view: in those markets, pay for a measured share of answer across the engines your buyers use, and treat citations as one input to that number rather than the thing you buy.

A decision you can make in one meeting

Start from the market you sell into. If answers in your category often link and your buyers click through to book or buy, a capped per-citation top-up on a fixed fee can make sense. If answers rarely link, or buyers decide on the names they read, buy a measured share of answer and leave citations as a diagnostic.

Then read the contract, not the pitch. The prompt list, the counting unit, the baseline and the cap decide what a per-citation deal costs, far more than the rate printed on the proposal. An offer that will not fix those four in writing is asking you to pay for whatever its tool reports.

Our view: "pay only when it works" is a fair request from a buyer, and a fair offer from a provider, once both sides agree in advance what working means. A measured share of answer on unbranded buyer prompts, counted the same way each month, is the closest thing this market has to that agreement.

Sources

  1. Similarweb, AI search stats 2026: ChatGPT answers with a citation, US, Jun 2025–May 2026 (Jul 2026)
  2. Ahrefs, ChatGPT query fan-out study: 66.6M free, logged-out GPT-5.6 Luna responses, Aug–Sep 2026, all countries; factual subset 20,000 US prompts; citation ranks US, Mar–Sep 2026 (Oct 2026)
  3. Semrush, 2026 AI Visibility Index: 126M US prompts, Jan–Apr 2026 (Jun 2026)
  4. Profound, where AI citations come from: 11.84B citations, 8 engines, 16 Apr–16 Jul 2026 (Jul 2026)
  5. Ahrefs, AI Overviews on branded searches: US desktop, Jul–Sep 2026; source shares and position from a separate panel of 100 brand-name queries, 14 weekly checkpoints (Oct 2026)
  6. Sigzen AI, published GEO prices: 34 of 41 agencies listed and 25 tools, seen 5–6 Oct 2026 (Oct 2026)
  7. MaxAEO, AI visibility sample size and clustered runs (Jul 2026)
  8. Seer Interactive, AI Overviews and CTR, 2026 update: 53 brands, Jan 2025–Feb 2026 (Apr 2026)
  9. Profound, commercial conversations in ChatGPT: 7.5M en-US conversations, 23 Jun 2025–22 Jun 2026, with industry splits (Aug 2026)
  10. G2, 2026 AI Search Insight Report: 1,076 B2B decision makers, North America, EMEA and APAC, Mar 2026
  11. Sigzen AI, published prices (Oct 2026)
  12. WebFX, How much does GEO cost in 2026? (May 2026)
  13. Tenten, 2026 GEO Pricing Benchmark, evidence reviewed 12 Aug 2026 (Aug 2026)
  14. SE Ranking, AI traffic study: 101,574 websites, 250 countries, Jan–Apr 2026 (Jun 2026)
  15. OpenAI, ChatGPT release notes: GPT-6 Sol for paid tiers from 7 Oct 2026 and GPT-6 Luna for Free and Go from 8 Oct 2026 (Oct 2026)

Questions readers ask

  • Pay-per-citation pricing for GEO work is a fee model in which an agency invoices a fixed amount each time an AI answer links to or attributes the client's site, usually with a small monthly fee or none. Its value depends on the contract: which prompts are counted, whether branded prompts count, how a citation is counted across repeat runs, and whether fees start only above a counted baseline.

Keep reading

Free. Enter your domain and Sigzen AI checks how five engines answer your buyers' questions, against up to three competitors.

No account needed. Two minutes to submit; we email the full five-engine report when the check has run.