Platforms are lying to you about performance. And you still believe them.

The platform grades its own homework. It gives itself straight A's

Let me start with an admission, because preaching is easy. For years I signed off client reports where ROAS came straight out of the interface, and it felt entirely normal. It took me a long time to notice how strange the arrangement is: an auction system assesses how much value it created, and is then awarded more budget on the strength of its own assessment. A supplier in any other category who wrote their own references would not survive the first procurement round.

Haus, a measurement firm specialising in geo experiments, published a summary of 640 tests on Meta. Two findings came out of it. Meta works as a channel, with an average lift in business outcomes of around nineteen percent. At the same time, fully automated Advantage+ campaigns beat manually managed ones on incremental ROAS for only two brands in five. Automation finds the people most likely to buy quickly, which means the people who were going to buy anyway.

A harder number comes from Wicked Reports, who analysed 55,661 campaigns. The cost of acquiring a genuinely new customer through Advantage+ rose from $257 in May 2024 to $528 in May 2025, doubling in twelve months. Reported ROAS barely moved, holding around 4.52. The dashboard stayed green while the acquisition economics underneath it halved.

What to do about it on Monday morning. A geo holdout test, meaning you switch the channel off in a share of regions and compare revenue against the rest, is currently the only method that buys you certainty. The average Haus test ran 18.6 days with a further 8.8 days of observation, just under a month in total. Smaller markets need longer runs to reach significance, but the principle holds. Switching Meta off in two regions once a quarter costs less than a year of funding campaigns that would have delivered the same result without you.

AI Overviews swallowed the clicks. Paid positions are paying for it too

This is where the model paid search has rested on for fifteen years starts to crack. BrightEdge measured AI Overviews on 48 percent of searches in February 2026, up 58 percent year on year. SparkToro adds that in 2026 fewer than a third of all Google searches send a click to any website at all. Seer Interactive then measured the most uncomfortable part: when an AI Overview appears on the page, click-through on paid results falls from 19.7 to 6.34 percent.

Read that again, because it is easy to skim past. This is not the organic traffic that SEO teams have been mourning for a year. This is a collapse on the positions you are paying for. The user gets their answer at the top, leaves satisfied, and your ad serves an impression that no KPI will ever count.

The second half of that finding is more interesting and almost nobody quotes it. Brands mentioned inside the AI Overview show materially higher click-through on their own paid positions than brands absent from it. A citation in the answer works like a recommendation from someone the reader already trusts: when the machine names you at the top, your ad further down suddenly looks credible. Paid search and generative engine optimisation have merged into one discipline, even though most companies still run them in two departments with two budgets.

The practical consequence for planning: track impression share specifically on queries where an AI Overview appears, and value those queries differently. Transactional phrases such as "buy" or "price" remain relatively untouched, because an answer does not put anything in a basket. Informational phrases at the top of the funnel lose value as a source of clicks and gain it as a source of citations. The same keyword now needs two strategies running side by side.

Creative is the last lever the algorithm cannot take from you

Once the platform took over targeting, budget allocation and placement, surprisingly little was left for the specialist. What remains is creative, the offer and measurement. Nielsen puts creative at roughly 47 percent of advertising's sales effect, rising to 56 percent in digital, which we covered separately. In PPC this shows up in one unexpectedly concrete way: formats that look like content consistently outperform formats that look like advertising.

On LinkedIn the gap is almost embarrassing. Standard single-image Sponsored Content sits around half a percent click-through in 2026 benchmark compilations, at somewhere between $5.74 and $6.50 per click. Thought Leader Ads, meaning paid amplification of a specific person's post, run around 2.68 percent click-through at $2.29 per click. Five times the click-through at under half the price is an anomaly in an auction environment, and the simplest explanation is probably the right one: people read people.

The same pattern holds on TikTok. Spark Ads, the paid amplification of a creator's organic video, convert at around 2.6 percent against 1.8 percent for studio-produced creative, with click-through of roughly 2.4 percent against one. A Spark Ad looks like a post you are used to seeing, while a studio spot looks like something somebody paid to put in front of you. Viewers tell the difference in about three tenths of a second, and the thumb is already moving.

Reddit grew by three quarters while the industry argued about whether anyone was there

The most interesting numbers of 2026 are not in an agency deck but in a quarterly filing with the US Securities and Exchange Commission, where misstating figures carries a prison sentence. Reddit reported $625 million in advertising revenue for the first quarter of 2026, up 74 percent year on year. Active advertisers grew by more than 75 percent. Performance advertising, meaning the lower funnel, now accounts for over 60 percent of its ad revenue. By growth rate it is the second fastest growing advertising platform in the world.

Two operational figures are worth writing down. The automated Reddit Max tool cut cost per action by 17 percent and raised conversions by 25. Dynamic product ads improved return on ad spend by 90 percent year on year. In two years Reddit moved from curiosity to a channel where routine acquisition works.

Read the last two rows carefully, because the whole question of timing sits in them. Volume and price both rose by 32 percent, so Reddit's cheap phase is closing. Enter this year and you pay twice what you would have paid last year, and considerably less than you will in two. On Quora, let us be honest: public data is thin, non-English volumes are small, and it works as a supplementary B2B channel for defending your brand inside answers that language models later read. The budget that belongs there fits into a single low four-figure line.

Local search engines shrank. Skipping them is still a mistake

Most global PPC guides write as though Google were the only search engine on earth. In a good half of the world's markets that is wrong, and the exceptions are commercially significant: Naver in South Korea, Yandex across parts of Eurasia, Baidu in China, Seznam in the Czech Republic. Our home market makes a useful case study, because the decline there is well documented. StatCounter put Google at 81.07 percent of Czech search in July 2026 and Seznam at 14.05, with the direction of travel unchanged for years.

Writing such an engine off would still be a mistake, for three reasons that have nothing to do with market share. First, price: thinner competition in the auction means a lower cost per click, so the channel's absolute economics tend to beat its size. Second, audience: domestic engines usually hold older and higher-spending users who convert well in specific categories. Third, inventory: Seznam folded its shopping comparison service into the main ad platform in April 2025, which means shopping campaigns, search carousels, image search and classifieds now run from a single interface. For a local e-commerce business it is the second most important shopping channel after Google Ads.

There is usually a measurement deadline attached as well. Seznam is currently rolling out a server-side event standard that replaces its old conversion and remarketing tags, in line with the wider move away from cookies. Advertisers who have implemented it will keep measuring. Those who have not will find a hole in next year's reports and blame seasonality.

X and TikTok: two very different bets filed under the same label

X is a rehabilitation case. Ad revenue has grown for five consecutive quarters and eMarketer estimates roughly $2.46 billion for this year, about fifty-eight percent of its pre-acquisition peak. In April 2026 the platform launched a rebuilt advertising system powered by Grok. Against that stands recent history: senior marketer confidence in X fell from 22 percent in 2022 to twelve percent in 2024, and a press release will not repair that. Buying X means buying cheap attention in exchange for adjacency risk you do not control. For a brand with a fragile reputation it makes no sense. For a brand that can afford an edge, the pricing is genuinely below market.

TikTok has the opposite problem. The cost of reach stays well below Meta, with various compilations putting cost per thousand impressions somewhere between four and five dollars against roughly fifteen for Meta. Conversion data, however, swings so violently that building a sensible benchmark from it is close to impossible, with one source reporting a year-on-year drop in conversion rate of more than a third. That spread is itself the finding: TikTok rewards creative production and punishes recycling, so your result depends on how much material you can make, not on how you configure the campaign.

Google Ads: still the largest, no longer the safest assumption

I left it until last deliberately, because it gets talked about the most and thought about the least. Google Ads remains the largest channel in the world and Performance Max is now its default shape. Independent measurement gives it a respectable rather than miraculous grade: median incremental return of around 4.6 with relatively low variance, meaning a format that behaves predictably. The problem is the distance between that figure and the report. Platform attribution overstates PMax contribution by roughly a third according to the same analyses, and for some advertisers reported return differs from actual return by a factor of two to five.

The overstatement has a mundane cause. PMax is handed brand search, remarketing and Shopping, then claims every conversion from all of them, even though most of those people already knew the brand. The defence exists and it is tedious: exclude brand queries, split remarketing into its own campaign, set real conversion values instead of unit ones, and verify the whole apparatus with a geo test once a quarter. None of that happens by accepting a recommendation in the interface, which is probably why almost nobody does it.

Then there is a layer that no PPC team can fix on its own. Binet and Field, working across 996 IPA case studies, arrived at a split of roughly 60 to 40 between brand building and activation, shifting to about 46 to 54 for B2B in their work with the LinkedIn B2B Institute. Benchmarks for 2026 show the median B2B budget running at seventy percent activation and twenty-five percent brand. In the same surveys marketers say they would prefer a very different split. They know it, they want it, and they still do not do it, because brand is hard to defend in a quarterly review.

Which closes the circle. Shifting budget into activation always looks better on a dashboard, because activation measures itself and brand measures slowly and awkwardly. The demand your brand creates, however, is exactly the demand that PMax and Advantage+ will later claim as their own achievement. Underinvest in it and you will eventually find your excellent performance channel harvesting an increasingly empty field. The dashboard will be the last thing to tell you.

Frequently asked questions

Sources: Haus, The Meta Report: Lessons from 640 Haus Incrementality Experiments · Wicked Reports, analysis of 55,661 Meta campaigns (June 2025) · Reddit, Inc., Form 10-Q for the quarter ended 31 March 2026 (SEC) · BrightEdge, AI Overview prevalence (February 2026) · Seer Interactive, AI Overview impact on CTR (2024 to 2025) · SparkToro, share of searches ending without a click (2026) · StatCounter, search engine share in the Czech Republic (July 2026) · Seznam, migration of shopping management and rollout of server-side event measurement · eMarketer, X advertising revenue estimates · Les Binet & Peter Field, The Long and the Short of It (IPA, 996 studies) and work with the LinkedIn B2B Institute · Nielsen, creative contribution to sales effect · compiled LinkedIn Ads and TikTok Ads benchmark reports for 2026. Treat figures from compiled benchmark reports as indicative: the spread between sources is wide and methodologies differ.

This is what we do: growth & performance marketing. Why creative decides the outcome is here, the brand to activation split here, and visibility inside AI answers here.

Suspect your campaigns report more than they actually deliver? We will work it out with you straight, and the first opinion is free. Get in touch.