Arbitrage is the practice of buying something in one market and selling it in another for more, keeping the difference. In advertising the traded good is attention. A click is bought from a search engine, a visitor from a content recommendation widget, or an impression from an exchange, and each is then resold to a second advertiser at a higher rate. The spread is the entire business; between the two transactions nothing is manufactured except the transfer.

The word is borrowed from finance, where an arbitrage means a risk-free profit taken because one asset carries two prices at once. Advertising arbitrage is not risk-free: the buy price is fixed at auction, while the sell price depends on click-through rates that may never materialise. The label stuck because the shape is identical, two prices for one unit of demand and an operator standing between them.

Three practices carry the name, and conflating them causes most of the confusion.

Traffic arbitrage and the maths of the spread

The best known version buys audiences and sells impressions. An operator purchases clicks from native networks, social feeds or search on a cost-per-click basis, routes them to a page dense with advertising, and earns on a cost-per-mille basis from whatever demand that page attracts. Profit is the gap between cost per visitor and earnings per visitor. At five cents in and fifteen cents out, a ten-cent margin looks trivial until multiplied across tens of thousands of daily sessions.

Making that margin work requires engineering the page rather than the content. Slideshow formats and paginated listicles turn one article into ten page views, while sticky anchors, autoplay video and stacked units raise the count per view. The same story served to a paid visitor is frequently broken into more pages, and carries more units, than the version an organic reader sees.

Costs on the content side have collapsed. DoubleVerify's Fraud Lab documented AutoBait, a network of more than 200 domains generating clickbait for roughly $2.25 an article with large language models and an image generator, after its operators left the prompts exposed in the sites' JavaScript. The network produced tens of millions of impressions, each domain presenting as an independent lifestyle blog. Production cost near zero changes the arithmetic: the spread no longer needs to be wide.

Inventory arbitrage inside the supply chain

The second version never touches a consumer. Intermediaries buy publisher inventory cheaply, often without authorisation, and resell it through exchanges at a markup. The explainer on ads.txt separates domain arbitrage, where a reseller misrepresents the source, from inventory arbitrage, where a middleman resells legitimate supply while taking a cut that never reaches the publisher.

The countermeasures are file-based. Ads.txt, published by the IAB Tech Lab in 2017, lets a publisher declare which sellers are authorised. Sellers.json and the OpenRTB SupplyChain object, finalised by the OpenRTB Working Group on July 31, 2019, let a buyer see every entity that sold or resold a bid request, each node naming a seller identifier that resolves against a sellers.json entry marked as publisher, intermediary or both. Centro made both files a requirement for its supply partners in 2020, citing duplicate impressions and bidding against itself.

Declaration is not verification. A February 2024 Pixalate analysis found 25% of programmatic traffic carrying SupplyChain data failed validation, with failing traffic showing invalid traffic rates 64% higher. The Trade Desk built its case for OpenAds on the same complaint, with chief executive Jeff Green naming duplication and obfuscation by resellers as the behaviour the platform was built to defeat.

Fees compound with each hop. The ANA's Programmatic Media Supply Chain Transparency Study, published in December 2023, traced 29% of spend to transaction costs before media quality losses, leaving 36 cents of each dollar entering a demand-side platform reaching a valid, viewable, non-MFA impression.

Origins: search arbitrage and the Google crackdown

Search arbitrage arrived with paid search itself. Operators bought low-cost keywords, sent traffic to thin pages carrying contextual advertising, and collected the difference. Google began closing the gap in 2005. Its Inside AdWords blog dated Quality Score and quality-based minimum bids to August that year in a December 2005 post, while an August 2008 post placed the change in July 2005. Either way, landing page quality entered the score in December 2005, and AdsBot began crawling advertiser destinations in 2006 to assess pages showing users the advertisements they had just seen.

Enforcement then became explicit. Google's Abusing the ad network policy still lists, among prohibited practices, "arbitrage" or promoting destinations for the sole or primary purpose of showing ads, alongside cloaking and bridge pages, while the Destination requirements policy disapproves landing pages built primarily to show advertising.

Geosign demonstrated the consequences. The Guelph, Ontario company took a $160 million investment from American Capital Strategies in March 2007, among the largest private financings for an internet company at the time, on a business built on click arbitrage across roughly 180 sites. Google's landing page enforcement arrived within weeks, costs rose, and the company was split between two owners by late 2007.

Parked domains outlasted it. Registrars and speculators monetised undeveloped addresses with related-search links to sponsored results, a category Google folded into the Search Partner Network as AdSense for Domains. Google removed them from the network on February 10, 2026, completing a gradual opt-out begun in early 2025 and deleting the setting from advertiser controls. Google separately began testing individual toggles for search partners and display in Performance Max in 2026.

The agency version

The third practice sits inside the buying chain. Under principal-based media, an agency buys inventory on its own account, takes the risk, then resells it to a client at an undisclosed markup. Trading desks made the model familiar in programmatic during the 2010s, and holding companies have expanded it since. Jeff Green used the term directly in March 2026, criticising firms that "arbitrage in the inefficiencies of programmatic" while presenting themselves as transparent, in a LinkedIn post following Publicis telling clients to stop using The Trade Desk after an audit dispute. PPC Land covered the exchange and the joint statement that closed it in June 2026 without resolving the underlying question of audit rights.

Why the practice matters

Arbitrage funds the made-for-advertising category, and the category is large. IAB Australia formally defined MFA sites in 2024 by auto-generated content, high ad density and paid traffic acquisition. The ANA's estimate that such sites take roughly 15% of programmatic spend and 21% of impressions is the figure most often cited against them.

Quality signals do not catch it. TAG, the ANA and Fiducia found that AI-generated inventory was graded premium more than 70% of the time, with lower invalid traffic than clean supply at 0.05% against 0.32%, higher viewability at 77.2% against 74.9%, and a higher price at $7.08 TrueCPM against $6.15. Arbitrage inventory is not cheap. It is expensive and it looks good.

Concentration is measurable by environment. Integral Ad Science recorded a 2.0% MFA rate on mobile web display against 0.5% on desktop during 2025, with mobile web generating 71.9% of all MFA impressions measured on the open web, and separately found traffic rising 5% across Christmas Eve and Christmas Day 2025, when campaign spending peaks.

Limitations and disputes

Whether arbitrage is a problem or a market function is contested. Every intermediary in programmatic buys low and sells high; the objection is to undisclosed spreads and to pages existing only to hold advertisements, not to margin itself. Google's policy draws the line at destination purpose rather than at profit.

The unit economics are also disputed. Practitioners including Augustine Fou have argued that traffic arbitrage cannot be profitable at realistic click-through rates without reliably delivered clicks, which points at non-human traffic as a hidden input rather than an incidental one. Measurement of the category meanwhile sits with vendors that also sell exclusion lists against it.

Direction of travel is unclear. The ANA reported MFA spend falling from 15% to 4% between 2023 and 2024 among participating brands, while generative tooling has since cut production costs to a few dollars an article, and ad fraud rates fell 41% in North America and 45% in EMEA during the first half of 2026 even as AI content volumes rose.

Disambiguation

Made-for-advertising describes the property; arbitrage describes the trade that fills it. An MFA site can also draw organic traffic, and arbitrage can route to a legitimate publisher.

Invalid traffic is non-human or fraudulent activity, defined and filtered under MRC standards. Arbitrage frequently rests on it, but buying real clicks and reselling real impressions is not fraud.

Domain spoofing misrepresents where an impression occurred. Arbitrage resells one that genuinely happened, at a place the buyer would not have chosen.

Remnant and reselling are authorised. Arbitrage becomes contested when the reseller is undeclared, or when the value added between purchase and sale is nil.

Recent developments

Detection moved to the model layer during 2026. DoubleVerify's AI SlopStopper blocked or monitored more than 500 million impressions across two regions in the first half of the year, while IAS moved its low-quality generative AI avoidance segment to general availability in May 2026 after opening a beta in April. Supply-path tooling has been automated in parallel, with IAS connecting post-bid transparency to pre-bid optimisation.

The trade keeps moving to whichever surface is least measured. Connected television, where fraud schemes rose 140% year on year in the first quarter of 2026, and mobile applications, where IAS blocked 812 domains in the Papyrus schemegenerating around $1 million a month, are the current destinations. The explainer on AI slop sets out the content supply behind the shift.

Timeline

  • July or August 2005: Google introduces Quality Score and quality-based minimum bids in AdWords, with company posts dating the change to both months
  • December 2005: Landing page quality becomes a Quality Score factor
  • 2006: AdsBot begins crawling advertiser landing pages to assess destination quality
  • March 2007: American Capital Strategies invests $160 million in Geosign, a click arbitrage operator
  • Late 2007: Geosign is split between two owners following Google's landing page enforcement
  • 2017: The IAB Tech Lab publishes ads.txt, allowing publishers to declare authorised sellers
  • July 31, 2019: The OpenRTB Working Group finalises sellers.json and the SupplyChain object
  • June 2023: The ANA reports MFA sites taking 21% of programmatic impressions and 15% of spend
  • December 2023: The ANA transparency study finds 36 cents of each dollar reaching a valid impression
  • 2024: IAB Australia publishes a formal definition of made-for-advertising inventory
  • February 10, 2026: Google removes parked domains from the Search Partner Network
  • March 4, 2026: DoubleVerify publishes the AutoBait investigation into a 200-domain AI clickbait network
  • May 2026: Integral Ad Science moves low-quality generative AI avoidance to general availability
  • July 28, 2026: TAG, the ANA and Fiducia find AI-generated inventory graded premium over 70% of the time

Summary

Who. Independent operators buying and reselling traffic, unauthorised resellers inside the programmatic supply chain, and agency holding companies trading media as principal. Google, the IAB Tech Lab, the ANA, TAG and verification vendors including Integral Ad Science and DoubleVerify define, measure and police the practice from different positions.

What. The purchase of advertising traffic, inventory or media at one price and its resale at a higher one, with the operator keeping the spread. It appears as traffic arbitrage on made-for-advertising pages, inventory arbitrage between exchanges, and principal-based buying between agency and client.

When. Present since paid search matured in the early 2000s, curtailed by Google's Quality Score and landing page enforcement from 2005 onwards, and reshaped since 2023 by generative tooling that cut content production costs to a few dollars a page.

Where. In search partner networks and parked domains, in the reseller nodes of the OpenRTB SupplyChain object, on mobile web display where made-for-advertising inventory concentrates, and increasingly in connected television and mobile applications where verification coverage is thinnest.

Why. Advertising prices vary by surface, format and buyer sophistication, and any persistent gap between two prices for the same attention invites someone to stand in the middle. The industry tolerates the margin and contests the disclosure, which is why arbitrage is simultaneously a prohibited practice in Google's advertiser policies and a growth strategy at listed holding companies.