媒体购买：比较价格前的四项检查

Media buying is the work of purchasing paid ad placements and managing them once they run. It covers how inventory is bought, what's negotiated or bid, and whether delivery, cost, and results match the deal.

Here's the fact that should slow you down: in large field experiments at eBay, brand-keyword search ads showed no measurable short-term benefit, according to the study's abstract (eBay paid search experiments, REPORTED).

So price should not be the first number you compare. Every media buy is really four agreements bundled together: the unit you're buying, how certain the supply is, which costs ride along, and what evidence will show whether the ads worked. Get one of them wrong and a cheap rate turns expensive.

Media Buying vs. Media Planning

Planning decides what to buy. Buying makes it happen. A plan sets the audience, the channel mix, the budget split, and how results will be judged, while buying turns that plan into actual orders, bids, deals, and delivery checks.

The two overlap more than org charts suggest. Planning doesn't stop when the first order goes live, because delivery data, cost surprises, and early results often send the plan back for revision while the campaign is still running. That's an editorial framing based on how the two functions hand work to each other, not a definition from a standards body.

Why this matters: if a team treats buying as pure execution, nobody owns the question of whether the purchase still fits the plan. And if a team treats planning as finished at launch, the buyer ends up rewriting strategy through bid changes without anyone deciding it on purpose.

How Advertising Inventory Gets Bought

The channel doesn't tell you the deal. Two display campaigns on the same site can be bought in completely different ways, with different guarantees and different pricing logic.

Guaranteed and Direct Deals

In a guaranteed deal, inventory is set aside for one buyer. Google Ad Manager describes Programmatic Guaranteed this way: "Inventory is designated only for that buyer at that price" (Google Ad Manager deal type comparison, product documentation, VERIFIED). The quantity is contracted, and the price is negotiated up front.

Notice the word programmatic in that name. Programmatic describes how the transaction is automated, not whether there's an auction. A Programmatic Guaranteed deal behaves much like a traditional direct buy; it simply runs through automated systems. That comparison is analysis drawn from the deal definitions, but it explains why treating programmatic and real-time bidding as synonyms hides a real difference in what you're promised.

Preferred Deals and Open Auctions

A Preferred Deal sits one step down. On the same Google Ad Manager page, the inventory "isn't reserved for the buyer," and "Buyers aren't required to buy the inventory" (VERIFIED). What the buyer gets is a first look at the negotiated price on each ad request.

That's a right of first look. It isn't a promise of volume.

Below that sit open auctions, where nothing is reserved and you compete impression by impression. How exchanges connect buyers and sellers is covered in the Ad Exchange glossary entry, and the auction mechanics are explained in how real-time bidding works.

Search and Social Auctions

Search and social platforms sell most of their inventory through their own auctions, and your bid is only one input. Google's ranking also weighs ad quality, thresholds, competition, and search context, which is why a higher bid can still lose, as the Ad Rank glossary entry explains.

Automated bidding changes what a "price" even means. With Target CPA, Google defines the target as the average cost per conversion you want to pay, and adds that "Some conversions may cost more than your target and some may cost less" (Google Ads Target CPA documentation, VERIFIED). You're setting an average to steer toward. You aren't buying conversions at a fixed rate.

Search inventory itself is shifting, too. As of October 2026, Google's help page on ads in AI Overviews says advertisers can't target placements inside AI Overviews specifically, can't opt out of them, and don't get segmented reporting when ads show there (Google Ads help on AI Overviews, VERIFIED). Indonesia is on Google's list of countries where ads inside AI Overviews run in English. For a buyer, that's a placement you pay for but can't isolate.

Teams running these auctions day to day usually split the work by platform, for example social media ads management versus search engine marketing campaigns. The four agreements apply to both.

What a Media Buy Really Costs

The quoted rate is rarely the full cost. Sometimes it isn't even the real rate.

Rate Card, Negotiated Price, and Bid

There are at least three different "prices" in play. A rate card is what a media owner publishes. A negotiated price is what a buyer actually agrees to, sometimes after a discount. A bid is what you offer in an auction, and with automated bidding it's an average target rather than a fixed charge, as the Target CPA documentation above shows.

Indonesia offers one documented example of why the distinction matters. Nielsen, a commercial measurement provider rather than a government body, says it monitors Indonesian advertising expenditure "at published rate card values," and notes that some media owners offer discounts whose rates aren't publicly available (Nielsen Indonesia advertising expenditure release, October 2025, VERIFIED for the methodology statement). That's a note about how one measurement service counts spend, not a portrait of the whole Indonesian market.

Worth checking before you compare two quotes. Is each one a rate card, a negotiated net price, or a bid target?

Media Cost vs. Billable Cost

Google's Display & Video 360 documentation breaks an invoice into parts. Media Cost is "the raw cost incurred for purchasing impressions." On top of that sit the platform fee (a percentage of media cost), platform add-on fees, third-party fees from vendors such as verification providers, country-specific surcharges where they apply, and taxes (DV360 fees and billable cost, VERIFIED). That breakdown is specific to DV360. Other platforms structure fees differently.

Here's the thing. Two buys with the same CPM can produce different invoices once platform, add-on, and third-party fees are added. Compare billable cost, not media cost, and know which line items your quoted rate includes. For how impression pricing is expressed, see the cost per mille pricing entry.

For scale, the market these line items sit inside is large. U.S. programmatic advertising revenue reached $162.4 billion in full-year 2025, out of $294.6 billion in total U.S. internet advertising revenue (IAB Internet Advertising Revenue Report, prepared by PwC; revenue, United States, 2025; REPORTED, unaudited industry association report). PwC states that it didn't audit the figures. Headline totals like these don't show what any single invoice contains.

Checking What Was Actually Delivered

Served vs. Viewable Impressions

An impression that was served isn't necessarily one anyone could see. Media Rating Council guidelines for mobile ads count a display ad as viewable when at least 50% of its pixels are in view for at least one continuous second after it renders, and a video ad when 50% of its pixels are in view with two continuous seconds played (MRC mobile viewable impression guidelines, final version June 2016, VERIFIED). If your deal prices served impressions, you may be paying for impressions that never met that bar.

So define the unit in the deal. Served or viewable? Then check delivery against that unit.

Who Was Authorized to Sell the Inventory

IAB Tech Lab standards answer a different question: who's allowed to sell. ads.txt lets publishers declare which companies are authorized to sell their inventory. sellers.json lets buyers see which entities are direct sellers and which are intermediaries. The SupplyChain object shows every party that sold or resold a given bid request (IAB Tech Lab sellers.json overview, VERIFIED). The same page notes that publisher fields are "inconsistently populated" in practice.

As analysis rather than a claim from the standards themselves: authorization isn't a quality guarantee. An authorized seller can still deliver ads nobody views. These standards tell you who sold the impression, not whether it was worth buying.

Measuring Whether the Buy Worked

Attributed vs. Incremental Results

Attribution assigns credit. It doesn't prove cause. A conversion that a platform attributes to an ad might have happened anyway, and the most reliable way to separate the two is to compare people who saw the ads against a credible group of people who didn't.

The eBay experiments by Tom Blake, Chris Nosko, and Steven Tadelis show why. According to the paper's abstract, brand-keyword ads had no measurable short-term benefit, and for non-brand keywords, frequent users whose purchases weren't influenced by ads accounted for most of the spend, which made average returns negative (NBER Working Paper 20171, published in Econometrica in 2015, REPORTED from the abstract). New and infrequent users did respond positively. These findings describe the eBay campaigns studied, not paid search everywhere.

Why Measurement Signals Keep Shrinking

The signals behind attribution reporting are getting thinner. Three changes stand out.

• Google is retiring its replacement APIs. In October 2025, after earlier deciding that Chrome would keep its existing third-party cookie choice approach, Google announced it would retire most Privacy Sandbox technologies, including the Attribution Reporting API, Protected Audience, and Topics, citing low adoption (Google Privacy Sandbox update, October 17, 2025, VERIFIED).
• Meta narrowed view-through windows in its API. Meta announced that from January 12, 2026, its Ads Insights API would stop returning 7-day view and 28-day view attribution windows, while 1-day view, 1-day engaged view, and 1-, 7-, and 28-day click windows remain (Meta for Developers announcement, VERIFIED).
• Safari blocks third-party cookies by default. That change dates back to March 2020, when WebKit announced full third-party cookie blocking in Safari 13.1 (WebKit's cookie blocking announcement, VERIFIED). It's older context, yet it still limits how much cross-site tracking a buy can lean on.

The reality is that platform reports now show less of the path, especially for ads people saw but didn't click. Planned tests matter more than reported totals.

Media Buying Mistakes That Look Like Savings

• Buying on CPM alone. A low CPM on served impressions can cost more per viewable impression once you apply the MRC standard and add platform and third-party fees. Compare the unit and the billable cost, not just the rate.
• Treating target CPA as a fixed price. Google describes target CPA as an average. Some conversions will cost more. Budget for the spread, not just the target.
• Reading attributed conversions as caused conversions. The eBay experiments showed how wide that gap can be in one retailer's campaigns. Ask what would have happened without the ads.
• Assuming an authorized seller means good inventory. ads.txt and sellers.json confirm who sold it. They don't confirm anyone saw it.

Media Buying FAQs

How is media buying different from media planning?

Planning decides the audience, channels, budget split, and how success will be judged. Media buying carries out those decisions through deals, bids, and delivery management. In practice the two overlap, because delivery results often send the plan back for changes.

Is media buying the same as programmatic advertising?

No. Programmatic is one way to buy, and it isn't limited to auctions. Google Ad Manager's Programmatic Guaranteed deals reserve inventory for one buyer at a negotiated price, while Preferred Deals give a first look at a negotiated price without any obligation to buy.

Is media buying the same as PPC?

No. Pay-per-click is one pricing arrangement within media buying. Media buying also covers negotiated deals, such as the Programmatic Guaranteed and Preferred Deals described above.

How much does media buying cost?

There's no single price. Cost depends on the deal type, the unit you buy, and the fees attached. In DV360, for example, the invoice combines media cost with platform fees, add-on fees, third-party fees, surcharges, and taxes. Published rate cards can also differ from negotiated prices.

How do you measure whether a media buy worked?

Start with delivery: did you get the agreed unit, such as viewable impressions? Then separate attributed results from incremental ones. Attribution shows how a reporting system assigned credit, while a test with a comparison group that didn't see the ads estimates what the ads actually caused.

Can you choose whether Google shows your ads inside AI Overviews?

Not as of October 2026. Google's help page says advertisers can't target AI Overview placements specifically, can't opt out of them, and won't get segmented reporting for them.

Related Terms

• Cost Per Mille (CPM)
• Frequency Capping
• Impression Share
• Programmatic Advertising
• Return on Ad Spend (ROAS)

Sources

• Google Ad Manager deal types, Programmatic Guaranteed vs. Preferred Deals, product documentation.
• Google Ads Target CPA bidding, product documentation.
• Google Ads help on AI Overviews, product documentation, accessed October 2026.
• DV360 fees and billable cost, product documentation.
• MRC Mobile Viewable Ad Impression Guidelines, Media Rating Council, June 2016.
• IAB Tech Lab sellers.json overview, standards documentation.
• Consumer Heterogeneity and Paid Search Effectiveness, Blake, Nosko, and Tadelis, NBER Working Paper 20171, 2014; Econometrica, 2015.
• Update on Privacy Sandbox technologies, Google, October 17, 2025.
• Ads Insights API availability updates, Meta for Developers, October 16, 2025.
• Full third-party cookie blocking, WebKit, March 24, 2020.
• Nielsen Indonesia advertisers release, Nielsen, October 14, 2025.
• IAB 2025 internet advertising revenue, IAB and PwC, April 16, 2026.