A glossary the ad industry hoped you wouldn't need

The marketplace where advertising space is bought and sold by machines. When a page loads, the exchange announces the available slot to bidders, runs an auction and declares a winner — all in roughly 100 milliseconds, before the page has finished drawing. Think stock exchange, except the commodity is your attention and the ticker never closes. The full choreography is walked through in The 100-millisecond auction.

The umbrella term for extracting advertising money without delivering real human attention: fake clicks, fake viewers, fake apps, fake sites. Because the supply chain is long and automated, fraud can hide at almost every link — bots that browse, clicks nobody made, plus a legal grey zone of made-for-advertising sites built for no reader at all. How much of the industry's money leaks away there is a story of its own — see Made for advertising — waste on a scale most industries would call a scandal, and this one calls overhead.

Google's advertising network for the long tail of the web: a publisher pastes a snippet of code into the page, Google fills the boxes with ads and shares the revenue. The ads arrive in cross-origin iframes, which means the host site cannot read, alter or even see them — a sealed window rented out in the middle of your own page. It is what funds this blog, as itemised in How this site makes money.

What you are worth, divided evenly: take a platform's revenue, divide by its users, and you have the going rate for a person. Meta's own reporting puts it at about $57 per person worldwide for 2025, and north of $200 a year in the US and Canada. It is the number that turns a 'free' service into a straightforward transaction — you are not the customer, but you are certainly on the invoice.

The observation that in a world of infinite content, the scarce resource is human attention — and whatever is scarce gets marketised. Feeds, notifications and autoplay exist because every second you spend looking is a second that can be auctioned; globally, digital advertising pays roughly $25,367 for each of those seconds (see The $25,000 second). 'Paying attention' stopped being a metaphor a while ago.

The art of deciding which ad deserves credit for a sale. If you saw a trainer ad on Tuesday, clicked one on Thursday and bought trainers on Saturday, attribution models exist to argue about who gets paid. The awkward truth is that the last ad you touched usually takes the prize — including for purchases you would have made anyway, which is why incrementality is the harder, more honest question.

The packet of information broadcast to advertisers the instant you load a page: the URL, your rough location, your device and browser, and whatever identifiers can be attached to you. A single page view can fan a bid request out to dozens or even hundreds of companies via real-time bidding, each invited to put a price on showing you something. Most lose the auction; all of them received the data. That asymmetry is the quiet scandal of the whole system.

Visits from software rather than people. Some bots are benign — search engines crawling, archivists archiving — but a portion exists purely to load pages and 'watch' ads so that someone gets paid for attention no human supplied. Networks filter what they can detect; what slips through is billed to an advertiser as if it were you. See ad fraud.

Advertisers' insistence on not appearing next to content that might tarnish them, enforced largely through keyword blocklists. The trouble is that a blocklist cannot tell reporting from endorsement: an article about a tragedy contains the word 'tragedy', so serious journalism goes unfunded while a celebrity listicle sails through. In practice, brand safety often means the news is too dangerous to pay for but a slideshow about yachts never is.

A neutral computing environment where two companies — say, a retailer and a platform — can match their customer records and measure overlap without either side, in theory, walking away with the other's raw data. Aggregate results come out; individual rows stay in. Clean rooms became fashionable at roughly the moment openly swapping user data became legally awkward — the handshake didn't stop, it just moved into a room with the curtains drawn.

Generating fake clicks on ads — via bots, click farms or malware — either to earn per-click payouts or to drain a rival's budget, since every fraudulent click bills an advertiser at the going CPC. Networks refund what they manage to catch. It is also why 'clicking the ads to support a site' is a genuinely bad idea: to the machine, enthusiastic charity and fraud look identical.

The software behind the cookie banner. A CMP records your consent choices and broadcasts them to the site's advertising vendors — frequently hundreds of them, listed somewhere behind a 'legitimate interest' toggle. Whether 'Reject all' sits beside 'Accept all' or three screens deep is a configuration choice, which tells you most of what you need to know; see dark pattern.

The machine's term for knowing nothing about you: a new user, a fresh browser, an empty profile. With no history to target against, systems fall back on context and broad demographic guesses until the data accumulates — which, given how quickly browsing piles up, is not long. Open a brand-new browser profile and you can watch it happen: for a brief, generic honeymoon, the ads treat you like a stranger. It doesn't last.

Google's mechanism for carrying on after you say no. When you decline cookies, tags running in consent mode stop setting identifiers; in its 'advanced' configuration they keep sending anonymised, cookieless pings, and the conversions that can no longer be observed directly are statistically modelled instead. Your refusal is honoured, in the sense that it is treated as a gap in the data to be estimated around.

Choosing ads based on what the page is about rather than who you are: golf clubs beside golf journalism, flour beside recipes. It is the oldest form of targeting and — with browsers and regulators turning against the third-party cookie — suddenly the future again. The distinction matters — a contextual ad knows what you are reading; a behavioural one knows what you read last month, at 2am, on a different device.

The thing the advertiser actually wanted all along: the purchase, the sign-up, the booked test drive. Impressions and clicks are only ever proxies priced against the probability of this moment, and an entire measurement industry — see attribution — exists to argue over which ad caused it. To you it was buying socks; to five intermediaries it was a billable event.

A pricing model in which the advertiser pays only when someone clicks — the norm for search ads, where a click signals intent. Auctions set the price, and intent is what costs: pennies for idle browsing, while clicks on searches that smell of lawyers or insurance can cost more than a good dinner. You are cheap to show things to; you become expensive the moment you seem ready to buy.

The base currency of display advertising: the price of a thousand impressions ('mille'), payable whether anyone clicks or not. Ordinary open-web display CPMs run to a few dollars, which prices a single glance from you at a fraction of a penny. There is a strange comfort in it — individually you are nearly worthless; only at scale do you become an industry.

Clicks divided by impressions: the share of people who saw an ad and did something about it. For ordinary display ads the figure is famously well under one per cent, which means the industry's definition of a successful campaign is one that more than 99 in every 100 people ignore. Much of the rest of this glossary exists to move that decimal point.

Interface design that steers you into choices you would not freely make: the luminous 'Accept all' beside the grey 'Manage options', the pre-ticked box, the cancellation flow shaped like a hedge maze. In advertising, dark patterns cluster around consent banners, where the difference between one click and five is measured directly in revenue. The tell is always the asymmetry — saying yes takes one tap; saying no takes a small hike.

A company that buys, aggregates and resells information about people it has no relationship with — you have never heard of most of them, and they have almost certainly heard of you. Brokers stitch profiles together from loyalty cards, apps, public records and other brokers, then sell membership of audience segments: 'expectant parent', 'payday loan seeker', or bleaker still — see In-market for a funeral. The product is you, pre-sorted.

The software advertisers use to buy. A DSP listens to bid requests streaming in from exchanges, decides in milliseconds whether you match a campaign's targeting, and bids accordingly; its counterpart on the selling side is the SSP. When people say 'the algorithm decided to show me this ad', a DSP is usually the algorithm in question.

The share of a publisher's ad slots that actually sell. A 90 per cent fill rate means one slot in ten found no buyer and gets a house ad, a charity filler or an empty box instead. If you ever see a website advertising itself in its own ad slot, you are looking at unsold inventory — the shopkeeper putting their own tins in the window.

Identifying your browser without a cookie, from its configuration quirks alone: screen size, installed fonts, timezone, graphics hardware and dozens of other traits that, combined, come surprisingly close to unique. Clear your cookies and the fingerprint remains — which is precisely the appeal. It is the technique of choice for recognising people who thought they had said no — the same instinct behind the hashed-email matching in The ads that follow you.

Information a company gathers about you through its own relationship with you: what you bought, browsed, clicked and volunteered at sign-up. As the third-party cookie falls out of favour, first-party data has become the industry's prized asset — hence the sudden enthusiasm for loyalty schemes, apps and 'create an account to continue'. Note that 'first-party' describes who collected it, not how modest they intend to be with it.

The minimum a publisher will accept for an ad slot: bids under the floor lose automatically, and the slot goes unsold rather than go cheap. Set it too high and the fill rate suffers; too low and the auction underprices the audience. So yes — even a moment of your attention has a reserve price, and somewhere a yield manager spends their working day tuning it.

A limit on how many times the same person is shown the same ad — three per day, say. Caps routinely fail across devices and platforms, because 'the same person' is exactly what the fragmented ad system struggles to agree on. When one sofa follows you from phone to laptop for a fortnight, the cap has not so much been broken as evaded; see The ads that follow you.

A publisher-side technique for making exchanges compete. Instead of offering a slot to one exchange at a time and settling for the first acceptable answer — the old waterfall — a script in the page header (Prebid.js is the standard) asks several at once and takes the best bid. It exists because publishers suspected, correctly, that going one buyer at a time was leaving money on the table.

The base unit of online advertising: one impression is counted each time an ad is served, whether or not a human ever saw it. Ads are priced per thousand impressions (CPM), which is why the machine cares a great deal about generating them and rather less about who is watching. An ad rendered at the bottom of a page you never scrolled to still counts — hence the separate, humbler metric of viewability.

An audience category for people the machine believes are actively shopping for something — "in-market for luggage", "in-market for SUVs" — inferred from searches, page visits and app behaviour. Landing in one raises the price of your attention, because advertisers pay more for someone about to spend. The inferences can be startlingly accurate, and occasionally bleak: see In-market for a funeral.

The measure of what advertising actually caused: sales that would not have happened without the ad, as opposed to sales the ad merely stood next to. Properly tested with holdout groups who see no ads at all — an experiment advertisers run less often than you might think, because the results can be humbling. The classic failure it exposes is retargeting claiming credit when you buy the boots you were already going to buy.

A targeting method in which a platform takes an advertiser's list of existing customers — the seed audience — and finds thousands or millions of strangers whose data resembles theirs. You can be targeted not because of anything you did, but because you statistically rhyme with people who bought the product. It is how an ad finds you before you have ever heard of the brand.

Sites built not to be read but to be monetised: recycled content, a dozen ad slots per screen, and traffic bought cheaply from clickbait widgets, all arbitraged against programmatic ad revenue. The ANA's 2023 transparency study found roughly 15% of programmatic spend landing on them. Perfectly legal, technically viewable, and the clearest evidence that the machine optimises for impressions rather than readers — the full story is in Made for advertising.

Ads designed to match the look, feel and format of the content around them — a promoted story in your feed, a "recommended" article at the foot of a news page. The whole point is that it does not look like an ad, which is also the whole problem. Regulators require a label; the label is usually the smallest, greyest text on the page. Its long-form editorial cousin is sponsored content.

Everything outside the walled gardens: the independent sites and publishers you reach through a browser rather than an app you are logged into. It is largely funded by programmatic advertising, which makes it both the messiest and the most inspectable part of the ecosystem — auctions between many parties leave paper trails that a closed platform never has to show. This site lives there, and is paid accordingly: see How this site makes money.

Data that identifies you as a specific person — name, email address, phone number. Ad tech's standard reassurance is that it does not trade in PII, only in "pseudonymous" identifiers: cookie IDs, device IDs, hashed emails. The distinction is real but thinner than it sounds — an identifier that follows one particular phone everywhere does not need your name to know you.

An invitation-only version of the programmatic auction: a publisher offers its ad space to a hand-picked set of buyers, usually at a negotiated floor price, over the same real-time bidding plumbing as the open market. Buyers get promises about where their ads will appear; publishers get better prices. It is the gated community of ad tech — same houses, higher fences.

The umbrella term for buying and selling ads by software rather than by people over lunch. Advertisers set targeting rules and budgets; machines execute them across millions of sites, most visibly through real-time bidding. Most of the world's digital ad money — a market spending roughly $25,000 every second — now flows this way, which is a lot of decisions for nobody in particular to be making.

The number of distinct people who saw a campaign at least once, as opposed to impressions, which count every showing. An advertiser can rack up a million impressions by showing one ad to the same insomniac a million times; reach is the metric that catches this, and the frequency cap is the tool meant to prevent it.

Showing you ads for something you looked at but did not buy. A pixel or cookie flags your visit; a DSP then bids to reach you, specifically, wherever you go next. It is why one glance at a pair of boots can commit them to following you around the internet for three weeks — anatomy in full at The ads that follow you.

The auction at the centre of it all. When a page with ad space loads, a bid request describing the slot — and, to varying degrees, you — is broadcast to would-be buyers, who bid; the winner's ad renders. The whole thing clears in roughly 100 milliseconds, faster than a blink, billions of times a day. Walkthrough at The 100-millisecond auction.

The starting list an advertiser hands to a platform — existing customers, subscribers, recent purchasers — from which the platform models a much larger lookalike audience. The seed is usually first-party data, often uploaded as hashed email addresses. Sign up for one brand's mailing list and you may quietly become the template for finding ten thousand people like you.

A brand's slice of all the advertising in its category: if half the trainer ads anyone sees are yours, your share of voice is 50%. Marketing folklore holds that brands whose share of voice exceeds their market share tend to grow — the theory that keeps rivals locked in expensive shouting matches whose surest beneficiary is whoever sells the ad space.

Articles, videos or posts paid for by a brand but produced in the publisher's own style — the long-form cousin of native advertising. Good publishers label it clearly and wall it off from the newsroom; less good ones count on you not noticing the word "partnership" in the byline. If a glowing article ends by recommending exactly one product, you have probably just read an ad.

The publisher's machine in the auction: software that offers a site's ad slots to exchanges and bidders, sets floor prices, and tries to extract the highest price for every impression. It is the mirror image of the DSP, which does the same job for buyers. Nearly every ad you meet on the open web has been haggled over by at least one of each, in the time the page took to load.

The same ad slot is typically offered for sale through many overlapping routes — different exchanges and resellers, each taking a cut. Supply-path optimisation is buyers pruning those routes down to the cheapest and cleanest. That it exists as a discipline tells you something: the plumbing grew so tangled that "work out where our money actually goes" became a job title.

The standardised filing system for people and content: the IAB publishes shared taxonomies so that a segment like "Interest | Travel | Cruises" means the same thing to every buyer and seller in the chain. It is how your browsing gets translated into tradable categories. Somewhere in these trees are labelled boxes with you in them — Google's My Ad Center will show you the ones it has filed you under.

A cookie set by a domain other than the site you are visiting — typically an ad server present on thousands of sites, which can therefore recognise the same browser everywhere and assemble a history of where it has been. It is the classic mechanism behind retargeting. Safari and Firefox block them by default; Google spent years announcing their execution in Chrome before deciding, in the end, to let them live. And where the cookie is blocked, fingerprinting is waiting.

The measure of whether an ad had a fighting chance of being seen. The industry standard, set by the Media Rating Council, counts a display ad as viewable if at least half its pixels are on screen for at least one second. Note where the bar sits: half the ad, for one second, is a pass. Ads that miss even that are still bought, sold and counted as impressions.

A closed platform — Google, Meta and Amazon are the big three — that keeps the audience, the ad system, the data and the measurement all inside its own walls. Advertisers get scale and precision; what they do not get is independent verification, since the platform selling the ads also marks its own homework. Everything outside the walls is the open web.

The old way of selling unsold ad space: offer it to one ad network, and if that network passes, cascade down a ranked list until someone takes it. Simple, slow, and prone to underselling, since the first taker won even when a network further down the list would have paid more. Header bidding — asking everyone at once — largely replaced it, which is roughly the moment publishers found out what their inventory was actually worth.