Generative Engine Optimization (GEO): what it is, how it works, and why the rules won't sit still

In March 2026, G2 surveyed 1,076 B2B software buyers. 51% now start their research with an AI chatbot instead of a search engine. 69% ended up choosing a different vendor than the one they had in mind when they started. One in three bought from a company they had never heard of before an AI recommended it.¹

Read those three numbers together and they describe a single event: the shortlist moved. It used to be assembled by a human, over days, from search results and referrals and whatever your sales team could get in front of them. Now, for half of B2B buyers, the first draft of the shortlist is written by a machine in the first thirty seconds of research. If the machine doesn't know you exist, or knows you but describes you wrongly, you are not losing deals. You are missing them, which is worse, because nothing shows up in your CRM to tell you it happened.

This guide is about that problem. It's called Generative Engine Optimization, GEO for short, and it is simultaneously one of the most important shifts in how businesses get found and one of the most carelessly explained. We run a GEO consultancy, so we have an interest here, and we'll be explicit about it when we get there. But most of this piece is the part nobody selling you a dashboard will say out loud: what GEO actually is, how the machinery works, why the tactics you read about last quarter keep dying, how to start doing it yourself this afternoon, and how to think about the space so the next platform change doesn't catch you flat.

What GEO is, and why it exists now

Generative Engine Optimization is the discipline of making your organisation visible, correctly understood, and recommendable to AI systems that answer questions — ChatGPT, Gemini, Perplexity, Google's AI Mode and AI Overviews, and the agents built on top of them.

The name is academic in origin. Researchers from Princeton and IIT Delhi coined it in a 2023 paper that treated "how do you get cited by a generative engine" as a measurable optimisation problem, and their findings still anchor the field.² We'll come back to what they found, because it kills a popular assumption.

Why does this discipline suddenly matter? Because in May 2026, the biggest distribution channel on the internet stopped being a list of links. At Google I/O, Sundar Pichai called the rollout of AI Mode "the deepest change in the history of the product": AI Mode passed 1 billion monthly users, and AI Overviews, the AI summaries above the classic results, passed 2.5 billion.³⁴ For the reader, that means the AI answer is no longer an early-adopter behaviour you can wait out. It is the default surface your buyers see first, on the default search engine of the world.

And it's not just the interface that changed. Google also announced information agents at the same I/O: AI that doesn't wait for a query at all. A buyer states an intent once, and the agent monitors the market for them continuously, surfacing candidates when something fits.⁵ Our read of that announcement, and we hold it as a company conviction: research is becoming ambient. The moment where you could "win the search" is dissolving into a standing evaluation that runs whether or not anyone typed anything. You are being evaluated on a Tuesday night with nobody at a keyboard.

Now put the supply side against that demand side. In April 2026, the 2X AI Innovation Lab tested how often B2B companies actually appear in AI-driven discovery: 96% were invisible, and only 4.3% showed up in early-stage buyer queries — the queries where shortlists get formed.⁶ It's a small study (70 companies), so treat the precise number gently, but the direction matches everything else in the field: Semrush's larger analysis found 62% of brands technically invisible to generative AI models.⁷ What that means practically: if you've done nothing deliberate about AI visibility, the safe assumption is not "we're probably fine." The safe assumption is that you are in the invisible majority, and your competitors' names are being spoken in conversations where yours isn't.

One more number, because it settles the "is this worth money" question. AI-referred visitors convert at roughly five times the rate of organic search visitors — 5.1× in an 18-month dataset from Rocket Agency.⁸ The mechanism isn't magic: by the time an AI-referred buyer clicks through, the AI has already done the comparing, so the visitor arrives pre-qualified and late-stage. The practical conclusion: AI visibility is not a vanity channel to monitor someday. It is currently the highest-intent traffic that exists, being divided among the 4% who show up.

A caution on reading those numbers, from a firm that lived them. Seer Interactive, a search agency that publishes its own performance data, watched 169,244 organic visits disappear from its site while AI-referred traffic replaced only about 7.5% of the loss. Then they noticed the part that matters: leads kept coming anyway.³¹ The clicks left; the value moved. What this means for you: if you measure this shift in traffic, you will misread it, possibly badly enough to cut the exact activity that's working. Fewer humans click through because the AI answers upstream, and the ones who still click are later-stage and worth more. The KPI that survives this transition is presence in the answer, and the leads that follow — not the visit count.

How the machine actually decides

Here is where most GEO content goes wrong, so let's be precise. When an AI answers "what's the best [your category] for [a buyer's situation]," several distinct mechanisms run, and they are optimised differently.

First, the model's memory. Large language models carry a compressed impression of the public internet from their training data. If your company has years of consistent, specific, third-party-corroborated presence, the model may simply know you, and can surface you with no live search at all. This layer moves slowly. You influence it the way you influence a reputation: over time, through what others publish about you. And it rewards patience — our read, and the reason we tell clients not to treat this as a quarterly campaign: models are periodically retrained on fresher snapshots of the web, so the consistent public record you build now is the one the next generation of models carries forward.

Second, live retrieval. For anything current or specific, the engine searches, reads sources, and synthesises an answer with citations. And it doesn't search the way a person does. Google's own documentation describes the "query fan-out" technique behind AI Overviews and AI Mode: one question gets decomposed into multiple related searches, issued simultaneously across subtopics and data sources, and the answer is assembled from everything that comes back.³⁹ What this means for you: there is no longer a single keyword to win. The machine probes an entire topic neighbourhood around the buyer's situation, which rewards businesses whose expertise is covered in connected depth and punishes the ones with one optimised page and nothing behind it. This is where GEO gets concrete, and where the founding research matters. The Princeton/IIT Delhi team tested nine optimisation methods across 10,000 queries: adding statistics improved a page's AI citation visibility by up to 41%, adding quotable expert statements improved subjective impression by 37%, citing sources helped by 30–40% — and keyword stuffing, the classic SEO reflex, had a negative effect.² (Those numbers are the researchers', not ours; you'll see them misattributed all over this industry.) What this means for you: the levers that win an AI citation are substance levers, not keyword levers. Machines cite pages that carry verifiable, specific, quotable information, because that's what makes a synthesised answer defensible.

Third — and this surprises almost everyone — the sources AI cites are mostly not your website. A July 2026 Search Engine Journal study ran AI agents across every major search surface and found 96% of citations pointed to third-party sources, up from roughly 80% just months earlier.⁹ Retrieval systems weight independent sources above brand-owned ones for an understandable reason: a claim about you from someone who isn't you carries editorial verification your homepage can't provide. The practical conclusion is uncomfortable and important: your AI visibility mostly lives on pages you don't control. A GEO strategy that only touches your own website is optimising the 4% and ignoring the 96%.

Fourth, the reading itself is stranger than you think. SEO rank and AI citation are already different games — BrightEdge found only 17% of sources cited in AI Overviews also rank in the organic top ten.¹⁰ It goes deeper. When Chrome supplies a page to Gemini, it doesn't send your page; a component called the Page Content Agent walks the rendered page and hands the model a structured tree of its readable elements. In one documented test, 471 extracted elements were filtered down to 198 before the model saw anything.¹¹ Your design, your hero animation, your carefully staged brand experience: discarded in transit. The machine reads a skeleton. If the skeleton doesn't say clearly what you do and who you're for, nothing else on the page gets the chance to.

Fifth, the machinery underneath rotates without notice. In July 2026, two independent analysts documented hidden source-selection pipelines behind ChatGPT's citations — internal systems with labels like Labrador and Bright that swap in and out week to week, changing which sources get cited while nothing about those sources changed.¹² For anyone tracking their "AI visibility score" weekly, this is the sentence that matters: some of what your dashboard reports as your performance is actually the plumbing rotating. Week-to-week citation movement is partly noise you cannot act on, and a strategy built on chasing it is a strategy built on static.

We compress all of this into a framework we use in every engagement, called the Three Gates. For an AI to send business your way, three things must be true in order: it can find you (discovery), it can read you correctly (comprehension), and increasingly it can act with you — book, quote, transact — on the buyer's behalf (capability). You can pass one gate and fail the next; the most common trap we see is the company that ranks beautifully for humans, gets retrieved, and then loses anyway because its information is a mess the machine can't lift cleanly, while a smaller competitor with cleaner data gets the recommendation. Each gate fails differently, and the fix for one does nothing for the others. Honesty requires saying that the third gate is early: the standards behind it are drafts, and we tell clients when it's not yet theirs to worry about. More on that below.

"But Google says you don't need any of this"

Before the market tour, deal with the strongest objection, because it comes from the platform itself. Google's official documentation states there are no additional requirements to appear in AI Overviews or AI Mode — no special optimizations, just the same fundamentals as regular Search.³⁹ Its guidance for succeeding in AI experiences reads like an SEO checklist with the volume turned down: unique, non-commodity, people-first content; clear structure; meet the technical requirements; good page experience.⁴⁰ Anyone selling GEO owes you a straight answer to this, so here's ours.

Take the guidance seriously — it's correct about what it covers. For Google's own AI surfaces, at the level of a single page, the fundamentals are the entry ticket, and Google's guidance even corroborates a number from earlier: it notes that clicks from AI results are higher quality, with visitors more likely to stay.⁴⁰ Anyone promising a clever trick that bypasses the fundamentals is selling you the exact behaviour the spam policies now name.²²

But read what the document is scoped to, because the scope is the answer. It's written by one platform, about that platform's surfaces, addressed to your website. It has nothing to say about ChatGPT, Perplexity, or Claude, which run their own retrieval machinery with their own rotating pipelines.¹² It has nothing to say about the 96% of AI citations that live on pages that aren't yours.⁹ It doesn't cover who in your organisation owns the problem when three engines describe you three different ways. And, tellingly, it doesn't mention that Google itself just shipped two new standards — ARD and OKF — whose entire reason to exist is that scraping websites isn't a good enough way for machines to understand organisations.²⁴²⁵ Our read, built on that evidence: "just do good SEO" is true advice about one gate on one platform, offered by the platform. GEO, done honestly, is everything the memo's scope leaves out.

The gold rush, and what happened to it

Once buyers moved, money followed. Fast.

Peec AI, a Berlin monitoring startup, went from launch to $10M in annual recurring revenue in 16 months and raised a $21M Series A in November 2025.¹³ Profound, the enterprise leader in AI visibility tracking, raised a $96M Series C at a $1 billion valuation in February 2026, eighteen months after founding.¹⁴ Adobe agreed to buy Semrush, the SEO suite repositioning around AI visibility, for $1.9 billion.¹⁵ Sitecore bought Scrunch, another monitoring platform, for a reported $225 million in June 2026.¹⁶ Beneath those headlines sit dozens of near-identical products launched in under two years, plus a long tail of hastily assembled dashboards wrapping the same handful of API calls — track some prompts, count brand mentions, chart a score.

Notice what almost all of that money bought: measurement. Dashboards that tell you where you appear. And notice what happened next, because it's the most instructive sequence in this industry's short history.

Alongside the dashboards, a second wave sold volume: AI-generated content at scale, hundreds of programmatically produced pages designed to blanket a category, listicles and comparison pages stamped out on autopilot. Entire platforms sell exactly this workflow today, openly, as a product.¹⁷

The cost of the volume playbook had already been demonstrated at brand scale, before the 2026 updates arrived. HubSpot spent a decade as the reference case for content marketing — and between November and December 2024, its monthly organic traffic fell from roughly 13.5 million visits to 8.6 million, with outside analyses putting the total decline at 70–80% of blog traffic within about a year.³² The post-mortems converge on the diagnosis: enormous volumes of broad, loosely related top-of-funnel content — famous-quote roundups, resignation-letter templates — written for a search reader who stopped existing, on exactly the queries AI now answers without sending a click, in exactly the format quality systems now demote. Two things make this case worth studying rather than gawking at. First, HubSpot wasn't bad at the old game; they were the best in the world at it, and it didn't protect them — when the reader changes, excellence at the old game transfers nothing. Second, HubSpot's own response to the coverage argued the business impact was far smaller than the traffic headline, because traffic was never the metric that paid the bills,³² which is precisely the Seer lesson from earlier,³¹ made by the casualty itself. Our read: what happened loudly to the most sophisticated content operation in SaaS is happening quietly, at smaller scale, to every business whose visibility strategy is still volume.

Then Google moved, twice, in the space of ten weeks. Its spam policies name "scaled content abuse" explicitly: using automation, including generative AI, to produce many pages without adding value for users.¹⁸ The policy is deliberately method-agnostic: Google's published guidance does not penalise AI-generated content as a category, it penalises unhelpful content however it was made.¹⁹ That distinction matters, and most commentary gets it wrong in both directions. The March 2026 core update enforced it at scale. The May 2026 core update, which began rolling out on May 21, the same morning as the I/O keynote, continued the pattern; some sites reported traffic losses above 50%.²⁰²¹ Our read of that timing, stated as a read: Google raised the AI answer layer and tightened the content quality floor in the same breath, because a generative engine is only as trustworthy as the corpus it draws from. Slop in, slop out. The engines have an existential interest in filtering low-effort content, which means the bar is not fixed. It rises with the technology.

Then, in May and June 2026, Google did something without precedent: it extended its spam policies to cover manipulating AI answers themselves. Buying citations inside AI responses. Content structured solely to trigger inclusion in AI summaries. Coordinated networks pushing specific sources into generative answers. All named, all now policy violations.²² For the reader, the significance is this: every "one weird trick" for AI visibility now has a regulator, and the regulator owns the surface. The gap between legitimate optimisation and manipulation stopped being an ethics debate and became written platform law — enforcement is admittedly hard, and researchers say detection is genuinely difficult,²² but the direction of travel is unambiguous.

So within about a year, the market produced: measurement tools that a billion dollars of capital is commoditising against each other, volume tactics that platform policy explicitly targets, and manipulation tactics that are now named spam. If you bought any of the three as "GEO," you bought a snapshot of a moving object.

Why this keeps happening: think in layers, not tactics

It would be easy to conclude that GEO is a scam because its tactics keep expiring. That's the wrong conclusion, and the pattern above shows why. The tactics expire because the ground is genuinely moving, and it moves on more than one level at once.

Consider what changed in a single eight-week window, mid-May to mid-July 2026, every item dated and public:

No tactic-level playbook survives that. But look at the list again and it sorts cleanly into four layers, and this is the frame we run our entire monitoring practice on:

1. Technology — the models themselves: capabilities, reasoning, retrieval behaviour. (New model releases change how sources get read and weighed.)
2. Infrastructure — protocols and standards: ARD, OKF, WebMCP, the agentic-commerce rails. (These define what "being visible" even means next year.)
3. The abstraction layer — the machinery between model and web: Chrome's page parser, the hidden source-selection pipelines, crawler access defaults. (This is what your dashboard can't see and your score silently depends on.)
4. Policy — platform and regulatory rules: spam policy expansions, core updates, what's compliant versus penalised. (This is what killed the volume play.)

These four move independently and on different clocks, and any of the four can invalidate a tactic on its own. That's our read of the evidence above, and it is the single most useful thing we can hand you from this whole piece: evaluate every piece of GEO advice by asking which layer it lives on and what happens to it when the other three move. A tip that only works while one pipeline is live, one policy is unenforced, or one standard is a draft isn't a strategy. It's a trade with an expiry date nobody printed on it.

A case study in fragility: Reddit

Nothing illustrates the layers better than Reddit, currently the most talked-about channel in AI visibility, and deservedly so.

The facts first. Users spent years appending "reddit" to Google searches to find human answers instead of marketing copy, and in May 2026 Google formalised the hint: AI answers now quote Reddit and forum threads directly.²³ Seer Interactive's analysis found ChatGPT citing Reddit in 13.5% of "judgment" responses — the which should I pick questions where buying decisions actually live — making Reddit 3–15× more cited than other social platforms.²⁷ Remember the earlier number: 96% of AI citations are third-party.⁹ Reddit is the loudest single voice in that 96%.

So yes: the unpolished thread where real customers discuss your category is now, quite literally, source material for your buyers' AI. Seer's own conclusion from the data, which matches everything we see in engagements: what earns citation is participation, not presence.²⁷ A named human from your company, in the conversation, being genuinely useful. A drive-by brand mention pasted by an agency does not read like that, to Reddit's moderators or to the models trained on what survives them.

Predictably, an industry of shortcuts appeared anyway: bought upvotes, seeded threads, aged accounts, keyword games riding Reddit's authority. Some of it even works, briefly, which is exactly the trap. Walk it through the layers and watch it die three separate deaths. Policy: buying citations inside AI answers is now named spam by Google, and Reddit's own moderation removes astroturf — you're gaming two referees at once.²² Abstraction: which pipelines feed Reddit content into which engines rotates without notice, so the lift you measure this month may be plumbing, not progress.¹² Technology: models get retrained, and content that got filtered as manipulation doesn't come back. Conductor's research already documents Reddit AI citations declining from their peak as engines recalibrate how much forum content to trust.²⁸

Our conclusion, framed as such: treat Reddit as a place your company's humans genuinely show up, or leave it alone. And treat the Reddit lesson as the general one: any channel that suddenly works this well will be gamed, then policed, then recalibrated. The participation survives every step of that cycle. The gaming doesn't.

Where this is heading: from mentions to verifiable authority

Everything above points one direction, and it's worth saying plainly because it changes what you should build this year.

The generative engines have a trust problem, and they know it — because researchers keep demonstrating how cheap the abuse is. Cornell Tech researchers showed that a planted snippet of user-generated text roughly thirteen words long can flip which product an AI agent recommends; the experiments ran in a controlled retrieval setup rather than on live platforms, but the retrieval mechanism is the one production agents use.³³ Microsoft's security team now tracks "AI recommendation poisoning" as an emerging attack class.³⁴ And a proper academic literature on adversarial GEO has formed: "preference manipulation attacks" that get an LLM to promote the attacker's product and discredit competitors, demonstrated against production AI search engines,³⁵ a 2026 benchmark built specifically to measure ranking manipulation in generative engines,³⁶ and a game-theoretic analysis with the bleakest finding of the lot — when manipulation works and goes unpunished, every publisher is incentivised to manipulate, degrading the shared information ecosystem until platforms are forced to intervene.³⁷ What this means for you cuts both ways. It means the machine's story about you and your competitors can be nudged by someone else, which is a reason to watch it. And it means the platforms, who have read the same papers, must build defences or watch their product rot.

Their answer, visible in every move of the past few months, is to narrow what they're willing to rely on: quality floors rising through core updates,²⁰ manipulation named as spam,²² third-party corroboration weighted over self-description,⁹ and, most telling of all, brand-new standards whose entire purpose is letting an organisation make verifiable declarations about itself. Agentic Resource Discovery is a machine-readable catalogue at a fixed, checkable address stating what your organisation is and offers; the Open Knowledge Format is your knowledge published in a structure agents read directly instead of guessing at your meaning from scraped fragments.²⁴²⁵ There is no keyword field to stuff in a declaration file. You either state, cleanly and checkably, what you are — or you leave the machine to piece you together from whatever it happened to scrape.

Honesty about maturity, because it separates advice from hype: these standards are weeks old and pre-1.0. A census the day after ARD launched found that of 39 major sites checked, including all eleven co-authoring companies, zero were serving the file.²⁹ So no, the deadline is not tomorrow. The reason to position early isn't panic; it's that these standards reveal where the platforms intend to take trust: away from "what can we scrape and hope" toward "what can be declared, corroborated, and checked."

One more measurement of how much ground this covers, because it's bigger than marketing. Gartner surveyed 3,566 customers in early 2026 and found them roughly three times more likely to take a service question to a third-party AI like ChatGPT than to the company's own chatbot — third-party AI use nearly doubled in a year while company-chatbot use has sat flat since 2022.³⁸ Put that beside the 96% third-party citation figure⁹ and the shape is hard to miss: before the sale and after it, the conversation about you now happens on ground you don't own. What remains ownable is whether you check out — wherever the conversation happens to be held.

Follow that trajectory a few years out — and we flag this as extrapolation, clearly labelled, not a sourced fact — and you arrive at a question worth sitting with: how far are we from an internet where AI systems simply discount whatever they cannot verify? Where being findable requires declarations that check out, corroboration you don't control, and a track record too consistent to fake? We don't know the date and neither does anyone else. But every dated fact in this piece moves toward that world, and none moves away from it. Which reframes the goal entirely: the game was never to sneak into an answer. You can still sneak into an answer this week; policy just made it spam, and retraining makes it temporary. The durable asset is authority a machine can check: who you verifiably are, what independent sources say about you, whether your claims hold when an agent pulls the thread. That's not a growth hack. That's reputation, made machine-legible. It compounds, and it doesn't rotate out with the plumbing.

How to do GEO yourself, starting today

None of what follows requires a consultant. This is the honest starting set, and for a lot of small businesses it's genuinely enough.

1. Ask the machines who you are. Open ChatGPT, Gemini, and Perplexity. Ask each: "What does [your company] do, and who is it for?" Then ask the question your buyer would actually ask: "What's the best [your category] for [your ideal customer's situation]?" Do it across engines, more than once — remember, citations rotate.¹² Twenty minutes, and you'll know your Three Gates position: absent (discovery problem), described wrongly (comprehension problem), or present and accurate. Most companies have never once run this audit on themselves.

2. Fix what the machine misread, at the source it read. If the AI gets you wrong, find where it got the wrong impression: usually your own vague homepage copy, a stale directory listing, an outdated third-party profile. Machines cite specifics; the founding research showed statistics, quotable statements, and cited sources lift AI visibility while keyword-stuffing hurts it.² Rewrite your core pages so a machine skimming the skeleton — headings, first sentences, plain statements — can answer what do they do, for whom, at what proof point without inference.¹¹ Google's own checklist for its AI surfaces is free and genuinely useful for this pass: people-first content, clear structure, technical requirements met, structured data that matches what's visibly on the page.⁴⁰

3. Check the plumbing. Confirm your site isn't blocking AI crawlers, deliberately or by default; since July 1, 2026, Cloudflare blocks several AI crawler categories on new domains out of the box.²⁶ Add structured data (schema) so entities — your company name, products, locations — are machine-stated rather than inferred. Google's new Search Console AI reports (announced June 3, 2026, rolling out gradually) show page-level AI Overviews and AI Mode impressions; note their limit, impressions only with no query data, so you can see that you appear but not why.³⁰ For a sense of the scale you might find: one large UK site measured 813,000 AI-surface impressions in a single week — roughly a tenth of its total search footprint, today, already.³⁰

4. Feed the 96%. Your AI reputation mostly lives on third-party pages,⁹ so behave accordingly: review platforms current, industry directories accurate, one or two genuinely substantive contributions where your buyers actually discuss the category — as named humans, participating.²⁷ One well-corroborated third-party mention is worth more to a retrieval system than another page on your own domain saying how great you are.

5. Watch the standards, calmly. Put ARD and OKF on a quarterly calendar reminder.²⁴²⁵ Pre-1.0 today; if your competitors are serving clean declaration files the quarter the standards mature and you aren't, that's the new version of not having a website in 2003. Early-and-calm beats late-and-panicked.

Do those five things and you are, without exaggeration, ahead of most of the market — the 96% who show up nowhere⁶ have typically done none of them.

When DIY stops being enough — and what we actually do

Here's the equally honest continuation: the five steps above are a snapshot, and everything in this piece argues that snapshots expire. The DIY playbook tells you where you stand today, against standards and machinery that we just watched shift six times in eight weeks. The hard problems start after the audit: why does the machine describe you the way it does, which of the four layers is your score actually sitting on, and who re-runs the judgement when the ground moves again next quarter?

That's the job CheckMarque AI was built for. A word on why, from our founder, because it's the honest origin of the whole company:

That last line is the conviction the whole practice runs on. The tools in this market overwhelmingly sell present-state measurement (some are genuinely good at it; we compare them honestly, by name, on a companion page). Numbers about today, unexplained. What they leave on the table is exactly the part this article has been about. The why behind every finding: which source, which gate, which layer, and what specifically to change on Monday. The whole organisation, because as the evidence above shows, machines synthesise you from everything — your site's skeleton, third parties, forums, structured declarations — which means AI legibility is an identity question that was never going to be solved from the marketing seat alone. And the road ahead: monitoring the four layers on a cadence and translating each shift into your answer ("this one's yours to act on, here's how; that one isn't, here's why") before it's mandatory, so you're never learning about a landscape shift from a headline.

The founding research made a subtler point that gets skipped in every summary: which optimisations work is context-dependent — the effective methods varied significantly by query domain.² Generic tools apply one playbook to every company by construction; that's what makes them scale, and it's also our read of why their advice flattens exactly where your situation gets specific. A fintech's trust gates are not an industrial manufacturer's. Bespoke isn't a luxury positioning claim here. It follows from how the machines actually behave.

If this piece did its job, you now know more about how AI reads companies than most of the people selling AI visibility. Run the twenty-minute audit. If what the machines say about you is wrong — or if you can't tell why it's right, or wrong, or about to change — that's the conversation we're for.

The machines are already describing you to your buyers, tonight, in conversations you'll never see. The only real question in GEO is whether what they say is up to you.