Launch Is a Day. Presence Is a Job.

Launch day is the best day. You ship. You post to Product Hunt. You write the Show HN. You answer every comment until 2 a.m., and for a week the graph looks like proof that this thing might work.

Then day 30 happens. Then day 200. Somewhere in there you renamed your pricing tiers, killed two features, and rewrote your tagline. Nobody updated the fourteen directories that still describe v1. Three of them are dead links now. A blog post from last year calls you “the new tool for X” — you pivoted off X in March. And when a founder asks ChatGPT for tools like yours, you aren’t in the answer. You were never in the answer.

Launch is a day. The other 364 days — the days when your product slowly disappears from every place you’re not actively maintaining — those are the actual job. Nobody’s job, currently. Which is the problem.

This article proposes that the job has a name: Presence Ops. Before you decide the name is silly, let me show you three things that quietly changed under all of us, one law that hasn’t changed at all, and a body of research that — assembled correctly — says something specific and slightly uncomfortable about how software gets discovered in 2026.

1. Most searches now end without a click

Rand Fishkin’s original zero-click study put the figure at ~50% of Google searches back in 2019. By 2024 it was 58.5%. In the 2026 update from SparkToro and Similarweb — based on January–April 2026 US panels — 68.01% of Google searches end with no click at all, the fastest increase in a decade. The open web now receives 22.9% fewer clicks per thousand searches than it did two years ago.

AI summaries are accelerating this. Pew Research analyzed 68,879 real Google searches and found that when an AI summary appears, only 8% of visits include a click on a traditional result — versus 15% without one. Ahrefs’ updated CTR study measured the #1 organic position on AI-Overview keywords falling from a 0.073 clickthrough rate in December 2023 to 0.016 by the end of 2025 — roughly 58% of the clicks Google used to send out, kept.

Here is the uncomfortable implication: on most surfaces, “being found” no longer means “getting the click.” The user reads the snippet, reads the answer, remembers a name — or doesn’t. The click was never the point of the long tail. Now it isn’t even the point of the head.

2. AI answers assemble from a wide pool — and the pool isn’t the top 10

When ChatGPT or Google’s AI Overviews compose an answer, where does the material come from? Semrush’s study of 200,000 AI Overviews found that an average answer cites ~11 sources, and those sources overlap with the traditional organic top-10 by only 20–26%. The #1 organic result appears in less than half of the AI Overviews for its own keywords. Google confirms the mechanism: query fan-out retrieves “a wider and more diverse set of links” than the blue-links era ever showed.

And 76–82% of AI Overviews occur on long-tail keywords — under 1,000 searches a month.

Meanwhile, in B2B specifically, the most-cited sources are not vendors’ own sites. G2’s aggregated research (drawing on Semrush, Profound, and Kevin Indig datasets) found that by October 2025 G2 had become the single most-visible domain in B2B AI answers — accounting for roughly a third of review-category citations in ChatGPT and AI Overviews, and up to ~75% on Perplexity. The page types AI engines cite are category pages and “best-of” lists. Customer reviews get quoted verbatim.

Read that again. The sources AI answers are made of are directories, lists, and review sites — third-party pages that describe you, not pages you control. Fewer than 7% of ChatGPT prompts produce any citation at all, which means being in the corpus those models draw from is the scarce event. Your listing on a DR-25 directory isn’t a traffic play. It’s an ingredient.

There’s one more finding worth its own paragraph. The Princeton GEO paper (published at KDD 2024) ran a 10,000-query benchmark and live experiments on Perplexity: adding citations, quotations, and statistics to source content lifted visibility in generated answers by 30–40% — and the gains were largest for low-ranked sources. The source sitting in position 5 gained +115%. In generative answers, the tail can beat the head. Domain authority is not destiny; being well-described somewhere real is.

3. Almost none of your buyers are shopping right now

John Dawes at the Ehrenberg-Bass Institute formulated what practitioners call the 95:5 rule: at any given moment, roughly 95% of your potential B2B buyers are out-of-market. They’re not comparing vendors this quarter. The job of marketing, in that model, is not conversion — it’s building the memory structure now, so that when the buyer enters the market in eight months, you’re the familiar name. The marketing-science lineage behind this is older and more quantitative than most growth blogs admit: excess share of voice above market share predicts share growth (Jones, HBR 1990; Binet & Field, IPA 2013), and Share of Search — a brand’s slice of category searches — explains roughly 83% of share of market across the IPA’s 30-case dataset, leading it by months.

Byron Sharp’s How Brands Grow gives the mechanism a name: mental availability — “the probability that a buyer will notice, recognize and/or think of a brand in buying situations.” It’s built through the quantity and quality of memory structures, and through category entry points: strong brands connect to many entry points rather than trying to monopolize one.

A directory listing nobody clicks is a category entry point. A best-of list you’re on, a review site profile, an accurate description in a niche aggregator — each is a tiny memory structure, in someone’s head or in a model’s training data, waiting for the day demand appears. Individually negligible. This is where the physics of the web comes in.

In 2000, Lada Adamic and Bernardo Huberman showed that website traffic, size, and links follow “a universal power law, characteristic of winner-take-all markets”. Two decades later, Ahrefs’ traffic study found 96.55% of pages in their index get zero Google traffic; 66.31% have zero referring domains. Attention is, and has always been, brutally concentrated in the head.

Chris Anderson’s “The Long Tail” (WIRED, 2004) argued you could “combine enough nonhits on the long tail and you’ve got a market bigger than the hits.” Anita Elberse’s famous HBR rebuttal (2008) showed the tail is also flat: on Rhapsody, the top 10% of tracks took 78% of plays, and 91% of tracks sold fewer than 100 copies. Anderson’s reply hinged on where you draw the head/tail line — but both sides agreed on the shape: a steep curve, a rich head, a poor tail.

Good. That agreement is the design spec. The head is where a few surfaces deserve heavy human investment. The tail is where per-surface returns are tiny — so tiny that manual work can never pay back — and only automation can push per-surface cost below per-surface return. This is the Head-and-Tail Rule:

• The head — Reddit, Hacker News, Product Hunt, the launch moments, the communities with real humans — is yours. Personally, manually, disclosed. Not because a vendor shouldn’t help, but because these surfaces punish anything less than authentic participation (ask the University of Zurich researchers whose 1,700 undisclosed AI comments got every account banned), and because your first users genuinely come from there. Every first-user retrospective says so. Handcraft the head. And note: the head isn’t outside Presence Ops — it’s the discipline’s most important ring, staffed by humans.
• The tail — hundreds of directories, tool lists, review sites, niche aggregators, coupon communities — is ops. No single one matters. Collectively they are your footprint in the corpus that both Google and the answer engines read. You cannot handcraft 300 surfaces. You can operate them.

And you can’t do the tail once, either, for two reasons. First: presence decays. Ward Cunningham’s original definition of technical debt — “shipping first time code is like going into debt… every minute spent on not-quite-right code counts as interest” — applies word for word to the listing that still describes your v1. Lehman’s first law of software evolution says used systems must be continually adapted or become progressively less satisfactory; your presence graph is a used system. Second: the tail regenerates. Google said in 2007 that 20–25% of each day’s queries had never been seen before. New directories appear, old ones die, new questions get asked in new places. The work is never finished — which is precisely why it’s operations and not a campaign.

Freshness isn’t optional overhead, by the way. Ahrefs analyzed ~17 million AI citations and found AI engines cite URLs averaging 25.7% fresher than organic search results; ChatGPT’s citations run 393–458 days fresher than Google’s, and fresher sources rank higher inside answers. The web page that was accurate when you launched and wrong today is not neutral. It’s a liability that compounds quietly.

The practice is a loop, run continuously:

1. Cover. Discover the surfaces where products like yours get listed, mentioned, and reviewed. The tail regenerates, so discovery never stops.
2. Verify. Every presence has a status: alive, indexed, accurate, dead. Most services hand you a spreadsheet once and call it done; the spreadsheet starts rotting the moment it’s delivered. Verification is the product.
3. Refresh. When the product changes, the graph changes with it — within days, not quarters. Presence debt gets paid before it accrues interest.
4. Measure. Coverage count, live-link rate, freshness lag, mentions in AI answers — a Presence Score with the same lineage as Share of Search: a leading, if imperfect, indicator that demand is being built before it arrives.

And two constraints that are architecture, not policy:

• Disclosed identities only. Official accounts, official submissions, official answers. The FTC’s reviews rule now carries civil penalties, and its recent enforcement reaches the agencies that fake engagement, not just the brands that buy it. Undisclosed presence isn’t a growth tactic; it’s an unpriced liability.
• You own the identities. Any service that registers accounts for you on its mailboxes owns your presence graph. Portability isn’t a feature; it’s the difference between renting and owning your footprint.

It is not link building. The goal isn’t PageRank or domain-authority arbitrage; Google explicitly states it mostly neutralizes low-quality links rather than rewarding volume. The goal is being accurately represented where humans and models look. A nofollow listing that keeps you in the corpus did its job.

It is not astroturfing. No fake reviews, no disguised consumers, no vote rings. Beyond the ethics and the FTC, it doesn’t even work for long: platforms ban it, and the ban takes your footprint with it.

It includes your launch. Product Hunt, the Show HN, the Reddit AMA — those are the head of your presence, its most important ring, and they stay human work: personal, manual, disclosed. The discipline covers head and tail. What starts where the launch ends is the delegable part — the tail, the 364 days. One discipline, two execution modes.

A theory that only cites its friends is marketing. So, honestly:

• Directory traffic is weak. Direct clicks from long-tail directories are, and will remain, modest. The value claim is corpus coverage, indexation, and AI-answer representation — not visits. Anyone who sells you the tail as a traffic engine is selling you 2011.
• Your first users still come from the head. Cold-community posts, DMs, launch events. The long tail is compounding infrastructure for the growth phase, not a spark for the first ten customers. If you’re pre-launch, go be human somewhere visible.
• The marketing-science numbers carry caveats. Excess-SOV effects were established on consumer-goods panels; Share of Search explains market share correlationally, not causally; B2B effects run slower (IPA’s own analysis says so). Presence Score should be treated the way good engineers treat any leading indicator: directional, watched over time, never worshipped.
• Cross-source consistency is best practice, not proven science. Keeping your name, positioning, and facts identical across surfaces feels obviously right and is what entity infrastructure (schema, knowledge graphs) expects — but nobody has run the large-N study proving it improves how models represent you. We do it anyway. We just don’t call it physics.
• The full chain is a hypothesis. Maintained state → corpus coverage → AI citation → business outcome: every link has evidence; no one has measured the whole chain end to end. We label this honestly — and we are running the experiment on our own products, with telemetry, before selling the result.

Anderson’s long tail was a claim about products: aggregate the niches and you rival the hits. Elberse showed the tail is flat and poor. Both were right, and the lesson for surfaces is the same as it was for songs. The head pays for craft. The tail only pays for operations.

What’s changed since 2008 is the payoff function. In a world where two-thirds of searches end without a click, where an answer is assembled from eleven sources you mostly don’t control, and where 95% of your market isn’t shopping yet — in that world, an accurate, alive, fresh mention on a small site nobody visits isn’t dead weight. It’s a bet that pays out the day a person finally searches, or a model finally answers, and you’re simply there. Described correctly. Current. Verifiable.

Launch is a day. Presence is a job. Jobs deserve disciplines, and disciplines deserve names.

We’re calling this one Presence Ops.

Sources are linked inline. The full research archive — including the claims we could not verify and chose not to make — is being prepared for publication under the same standards. Corrections: see the policy.