Most of the citations AI engines hand out are won on pages you do not own — and most PR was never designed for how engines read. Industry tracking finds roughly 43% of brand mentions vanish when an AI summarizes the article they sit in: the placement existed, the citation never happened. Citation engineering closes that gap. We place your brand in the sources ChatGPT, Claude, Gemini, Perplexity and Google's AI answers already cite for your category, structured so the mention survives extraction — and we measure the result per engine, not in placement counts. English and Spanish, fixed quarterly pricing below.
Across every credible 2025–2026 study, the signals that decide who gets named in AI answers live mostly on third-party pages. Ahrefs' analysis of 75,000 brands found branded web mentions correlate with AI visibility at 0.664 — the strongest factor measured, roughly three times backlinks at 0.218. Muck Rack's May 2026 study attributed 84% of AI citations to earned media, with paid content at a rounding-error 0.3%. And the channel is genuinely separate from rankings: Moz's analysis of 40,000 queries found 88% of Google AI Mode citations do not appear in the organic top ten, while a 2025 academic study found over a third of AI-cited domains absent from traditional search entirely.
The most direct experiment to date measured causation rather than correlation. In March 2026, Stacker and Scrunch tracked 87 earned stories across 2,600 AI prompts and recorded a median 239% lift in brand citations when the same content moved from owned channels to third-party publications — same substance, different surface, more than triple the citations. The mechanism is the consensus-building described in our six-step method: an engine deciding whom to name leans on corroboration across sources it trusts, and a claim that exists only on your own domain gives it exactly one source, which is to say none.
Correlation with citation (0 = none, 1 = perfect). Source: 2025–2026 studies correlating ranking and off-page signals with AI-Overview and chatbot citations.
Two honest caveats, because the numbers above get quoted carelessly. The earned-media share of citations varies with how you define it — studies using a strict definition report figures closer to a quarter, while broader definitions reach the eighties — so treat any single percentage as a directional reading, not gospel. And correlation is not a lever you pull once: mentions decay, engines refresh, and the brands that hold citation share are the ones still earning mentions next quarter. Both caveats are arguments for measurement, which is why it is a phase of this program rather than a promise in this paragraph.
The budget implication deserves saying out loud, because it inverts a decade of habit. Most marketing budgets still weight link acquisition over mention earning by muscle memory, yet the correlation data runs three to one the other way, and the experimental data shows distribution multiplying citations where owned publishing alone plateaus. If your AI visibility line item exists at all, the evidence says its center of gravity belongs off-site — in the mentions, descriptions and corroborations that engines actually weigh when they decide whose name to say. On-site work makes you quotable. This work makes you the brand that gets quoted.
Here is the failure mode traditional PR reporting cannot see. An agency lands you a mention in a respected publication and reports the placement as a win. Then a buyer asks an AI engine about your category, the engine retrieves that very article, summarizes it — and your name does not survive the summary. Connective3's analysis of more than 3,500 digital PR links across 170 brands found this happens to roughly 43% of brand mentions. The placement was real. The money was spent. The citation never existed.
The mentions that survive share a structure, and it is engineerable: the brand name in the same self-contained passage as the claim and the evidence, early in the piece, inside substantive context, consistent with how the rest of the web describes the company. That structure is what we brief, shape and negotiate into every placement — which is why this service is named citation engineering and not PR. The deliverable is not coverage. It is a mention an extraction pipeline keeps.
There is a second discipline inside the program that costs less than placements and often pays faster: correcting what already exists. Engines repeat what their trusted sources say, so when a comparison article lists your pricing from two years ago or files you in the wrong category, that error compounds into every answer built on it. Getting the source page fixed — a correction request, an updated quote, a refreshed listing — rewrites what every downstream engine reads. In most first quarters, two or three source-level corrections move the description-accuracy numbers before a single new placement lands, which is why corrections are a standing line of the program and not an afterthought.
Every placement in the program is held to four requirements. Miss any one and the odds of surviving extraction drop sharply.
placement
The mention lives on one of the pages your baseline shows the engines actually citing for your category — a comparison article, review platform, community thread or trade publication. Presence in a publication no engine reads is publicity, not citation engineering.
structure
Your name sits in the same self-contained passage as the claim about you and the number or fact that supports it. When the engine extracts that passage, everything travels together. Separate them across paragraphs and summarization keeps the claim while dropping the name — the 43% failure mode.
position
Mentions in the opening third of an article and inside substantive context get extracted disproportionately. A name-drop in a closing roundup paragraph reads as filler to a model the same way it does to a person.
coherence
The mention describes you the way your entity data does — same category, same framing, current facts. Contradictory descriptions across sources make models hedge; coherent ones compound. This is why citation engineering chains naturally after entity work.
typical mention · fails extraction
"…The market has matured considerably this year. Several vendors now offer credible options, including Acme, BrightStack and others, each with different strengths depending on team size…"
The claim is generic, the evidence is absent, and the brand sits in a list. When a model summarizes this passage it keeps "the market has matured" and drops the names — this is the 43% in action.
citation-grade mention · survives
"…For mid-market SaaS teams, Acme is the strongest fit in this group: its published benchmark across 1,200 deployments showed a 31% faster onboarding time than the category median, and it is the only vendor here with transparent pricing…"
Brand, claim and evidence travel in one self-contained passage, early and specific. A model extracting this keeps the name, because the name is load-bearing.
Both mentions could come from the same article and cost the same effort to earn. Only one of them exists after summarization. The difference is not luck or budget; it is a brief — and the brief is the product.
Traditional PR targets prestige: the biggest publications a budget can reach. Citation engineering targets what the engines demonstrably read. When you run a category's buyer prompts across five engines and collect every cited URL, the result is almost always a concentrated list — typically twenty to forty pages doing the majority of the citing, a mix of comparison articles, review platforms, a community thread or two, and a handful of trade publications. That list is the source map, and it changes the economics of earned media completely: instead of pitching a hundred outlets hoping some matter, you work a short list where every success is, by construction, on a page the engines already trust.
The map also ranks itself. Each source carries a citation frequency from your baseline, so the program works down the list in order of demonstrated influence rather than gut feel. And it exposes the asymmetries worth money: the mid-tier comparison site every engine cites for your category is worth more to your AI visibility than a prestige masthead none of them retrieve — a reversal of PR instinct that the data forces, and one reason placement counts and citation results diverge so badly.
The map is rebuilt every quarter, because it moves. Sources gain and lose the engines' trust, new comparison pages appear, and a community thread that dominated retrieval in March may be stale by September. A standing source map, refreshed against your own prompts, is itself an asset most competitors do not have — and it compounds, because each quarter's measurement teaches the next quarter's targeting.
01
From your baseline prompt set, we extract the actual pages each engine cites for your category — typically a concentrated list of twenty to forty URLs — and score them by citation frequency, accessibility and realistic path to presence.
02
We earn citation-grade mentions on mapped sources through original data, expert commentary and contributed analysis, and we correct existing descriptions of you that are wrong at the source — often the fastest win on the map.
03
The same baseline prompts, re-run per engine, with citation rate, share of answer and description accuracy reported against the quarter’s placements. You see what moved and what did not, attributed honestly.
The raw material matters as much as the targeting. Mentions are earned with substance — original data from your operations, expert commentary your team can genuinely stand behind, contributed analysis that gives an editor a reason to publish. Original numbers pull double duty here: 2026 benchmarking found content built on original data earns AI citations at roughly three times the rate of recycled statistics, and a study only you could have run makes you the only possible source for it. What we do not do is wire-blast press releases or buy placements; paid content earns 0.3% of AI citations, and engines discount promotional tone wherever it appears. The cadence stays deliberately small and finished: a handful of citation-grade mentions that survive extraction beat a spreadsheet of placements that do not, and the quarterly report shows you both numbers so the difference is never an argument, only a reading.
Citation engineering pays best for B2B SaaS and service brands in categories buyers research through AI — especially where a handful of comparison pages and review platforms dominate what engines cite, because a concentrated source map is a workable source map. It is steps five and six of the six-step method, and it assumes the earlier steps: if your own pages are unreadable to crawlers or your entity is ambiguous, mentions will point at a brand the engines cannot verify, and you will pay earned-media prices for entity-problem results. In that case start with the entity and schema sprint; we will tell you the same in the audit, for free.
A note on category dynamics worth weighing before you start. In categories where the source map is dominated by two or three review platforms, presence there is close to mandatory and the program leans toward profile depth, review velocity and the comparison pages those platforms publish. In categories where communities carry the citations — developer tools are the classic case — the work shifts toward being usefully, honestly present where practitioners actually talk, which cannot be faked and takes longer. The source map tells us which game your category is playing before we commit a quarter to it, and we will show you that map in the snapshot rather than asking you to take the strategy on faith.
It is also premature for brands with nothing citable to say. Mentions need substance to carry, and a company with no data, no point of view and no expertise on record gives an editor nothing and an engine less. The fix there is usually one good original study before any outreach — often a thing we build together in the first quarter. The AC Group has earned attention online for 27 years, and the lesson of every era has been the same: distribution amplifies substance, it cannot replace it.
citation engineering · quarterly
€4,800/quarter
One language. Source map from your baseline, citation-grade placements and source-level description corrections, per-engine measurement against the baseline prompt set.
Start a quarterbilingual · EN + ES · quarterly
€7,200/quarter
The full program across English and Spanish source ecosystems with one unified entity narrative — where the thinner Spanish source map usually makes each mention count for more.
Start bilingualQuarter to quarter, no annual lock-in: the measurement phase exists precisely so you can decide renewal on evidence. If a quarter's numbers do not justify the next one, we would rather you stop — a client who renews on data renews for years, and that is the only kind we want. Placements are never paid media, and corrections to existing wrong descriptions are included, not billed as extras.
Citation engineering is earned media designed for AI extraction: getting your brand mentioned in the third-party sources AI engines already cite for your category, with the mention structured so it survives summarization — your brand, the claim and the supporting evidence in the same extractable passage. It differs from traditional digital PR, which counts placements; research shows roughly 43% of brand mentions disappear when AI engines summarize the articles they appear in, and engineering the mention is how you stay in the surviving 57%.
Three ways. Targeting: we work from your AI visibility baseline — the specific pages engines cite for your category prompts — rather than a generic tier-1 media list. Craft: we brief and shape mentions so brand, claim and evidence sit in one extractable passage, the format that survives summarization. Measurement: success is your citation rate per engine before and after, not a placement count. A placement that AI summarization strips your name from is a cost, not a result, and standard PR reporting cannot see the difference.
Yes — they are the strongest signal measured. Ahrefs’ study of 75,000 brands found branded web mentions correlate with AI visibility at 0.664, roughly three times the 0.218 correlation of backlinks. Models read text, not link graphs: a plain-text mention of your brand in the right paragraph of a trusted comparison article does real work whether or not it links to you.
No, and you should walk away from anyone who does. Editorial decisions belong to publishers, and citation decisions belong to the engines, which also refresh constantly. What we control is the inputs the evidence says matter most: presence in the sources engines already trust, mentions structured to survive extraction, and a consistent entity behind them. What we commit to is measurement honest enough that you can see exactly what moved, per engine, per quarter.
Retrieval-based engines can reflect a new mention within days to weeks of publication — Perplexity usually first. Engines that lean more on periodically refreshed knowledge take one to three months. A quarterly cycle exists because that horizon is where the compounding becomes visible: the 2026 Stacker and Scrunch research measured a median 239% citation lift from distributing earned stories, accruing over weeks rather than overnight.
Yes, and the opportunity is usually larger. Spanish-language source ecosystems are thinner, so a well-placed mention in a trusted Spanish comparison site or trade publication carries proportionally more weight, with less competition for it. We run citation engineering natively in both languages with one unified entity narrative, which is rare enough among providers to be an advantage by itself.
The free AI visibility snapshot maps your category's citation landscape across five engines — who gets named, from which sources, and where you stand. It is the source map this program starts from, and the honest way to size the opportunity before spending a euro on it.
