A drip falling on dark stone above a thick layered mineral crust, with a shallow wet ring drying nearby.

What compounds in marketing and what resets every month

PR, content and citations accrue while paid resets at zero on the first. A budget split is really a decision about how much of your spend has a memory.

Turn off your paid campaigns tonight and tomorrow's traffic from them is zero. Turn off your content and PR tonight and the traffic keeps arriving for months, decaying slowly, on work you already paid for.

That difference is the most important thing in a marketing budget and it almost never appears in the budget document.

The two kinds of spend

Spend with a memory. Content that ranks and gets cited. Earned media that stays on a masthead. Reviews. Backlinks. Video that keeps being served. A branded search habit. Each unit of work adds to a stock that is still there next year.

Spend without one. Paid search, paid social, display, sponsorship activations, most events. The moment the money stops the delivery stops. Excellent for controllability, terrible for accumulation.

Neither is better. They do different jobs. The mistake is running a business where 90 per cent of spend has no memory and then wondering why the cost of acquisition rises every year.

The honest case against content

Content compounds only if it is worth compounding, and most of it is not.

SE Ranking ran the cleanest test available: 20 brand-new domains, 2,000 AI-written articles, no human editing, no links, 16 months. About 71 per cent indexed in the first month and the sites pulled over 122,000 impressions. By month three only 3 per cent of pages were still in the top 100. The August 2025 spam update briefly lifted that to 20 per cent, which did not hold. Nearly all impressions landed in the first two and a half months.

So it indexed, it spiked, and it died. That is not a compounding asset, that is a firework.

Now the counter-intuitive finding. Ahrefs analysed one million AI Overview results and 1.9 million cited URLs in July 2025. Of the pages being cited, 3.6 per cent were pure AI, 8.6 per cent pure human, and 87.8 per cent mixed. The correlation between how much AI was in a page and its citation order was 0.017, effectively nothing.

Google is not detecting and demoting AI at the citation layer. It is demoting content with nothing in it. Google's own guidance names the failing format directly, giving the example of commodity content like a listicle of tips for first-time buyers.

The dividing line is not who typed it. It is whether the page contains something that did not exist elsewhere: your data, your project, your method, a number only you have.

What Google says about the AI-search shortcuts

Google published a guide on optimising for generative AI features in May 2026, and it kills several things being sold right now.

llms.txt files: Google Search does not use them. Structured data: not required for generative AI search. Breaking content into tiny chunks for language models: no requirement to do it. Manufacturing brand mentions across the web: not as helpful as claimed.

The one hard gate is boring. The page must be indexed and eligible to appear with a snippet. If you have suppressed snippets anywhere, you have suppressed AI eligibility on those pages.

The volume question

AI-referred visitors convert far better than search visitors and there are not many of them. Semrush puts AI-referred conversion at roughly 4.4 times organic. Seer's client data shows ChatGPT-referred traffic converting near 16 per cent against about 1.8 per cent for Google organic. ChatGPT referrals grew 206 per cent year on year to January 2026, and still sit at around one per cent or less of most sites' sessions.

Small, high quality, growing fast. Plan for it as a two-year line, not a this-quarter line.

A split that survives contact with a board

For a mid-sized business with an existing brand, a workable starting shape is roughly two thirds of spend on the line that pays this quarter and one third on the line that builds the stock. Then move the ratio based on one question: is your cost per acquisition on paid rising year on year? If it is, your stock is too small and the paid channel is doing work that brand should be doing.

Three rules that make the compounding third actually compound.

  1. 1.Publish fewer things with more in them. One piece a month containing your own data beats twelve summarising other people.
  2. 2.Update rather than add. A page refreshed with this year's numbers usually outperforms a new page on the same topic, and it keeps the links it earned.
  3. 3.Count citations, not just rankings. Track whether your pages are being quoted inside AI answers on your commercial queries, because that is the difference between about 2.1 per cent and about 0.9 per cent click-through on the same result.

The trap

Compounding assets are slow, and slow work is the first thing cut when a quarter looks tight. Every cut resets the clock, and the businesses that cut it twice end up entirely dependent on the channel that charges them more every year.

Decide once how much of your spend is allowed to have a memory, and defend it in the quarters when it is hardest.

Written by David Eid. Published .