The short version: With the free foundation in place, a modest ad budget speeds things up rather than propping up a weak base. The cheapest click is not automatically the best click — what matters is intent. Google captures people already searching for what you sell; social interrupts people to create demand. For a considered purchase like a week of frozen dinners, captured intent usually converts better, so the spend order matches the rest of this guide: Google first, social second as a demand-builder.

Why you do not start here

It is worth being blunt about where this chapter sits. Advertising is Chapter 5, not Chapter 1, on purpose. A paid budget poured onto a weak foundation is money lit on fire — you pay to send a customer to a product page that does not convince them, a checkout that fails them when they turn out to be out of zone, or a brand with no reviews to reassure them. The ad spend amplifies whatever is already there. If what is already there does not work, you are simply paying to discover that faster.

So the first move in this chapter is a gate, not a tactic: confirm the free foundation is live before you spend. Chapters 1 to 3 — a clean Merchant Center feed, answer pages that convert, a checkout built for frozen, and genuine reviews — must be working first. Once they are, a small budget does what it is supposed to: it accelerates a machine that already runs.

One myth to clear up first

A belief that costs frozen merchants real money: that you must spend on Google Ads before your organic ranking improves. That is not how it works.

Paid ads do not directly raise your organic ranking. Google states this plainly, and you can rank well with zero ad spend — the free foundation of Chapters 1 to 3 is exactly that. But paid and free do work as a team: ads buy you visibility today while organic and AI visibility build over months, and ad-driven traffic generates the orders, reviews and engagement that indirectly strengthen everything else. The honest framing is that a small budget accelerates the free engine; it never replaces it, and you never need it to start.

Where to spend it first: intent beats cheapness

Here is the decision most guides get wrong by chasing the lowest cost-per-click. The platforms are not interchangeable, and the cheapest click is not the best click. The distinction that matters is intent.

Google Search and Shopping — and therefore Performance Max — capture existing intent. Someone typing “frozen meals delivered Sandton” or “low-carb frozen dinners Cape Town” is most of the way to a decision. They have the problem, they are looking for the solution, and you simply need to be there. This is the warmest traffic you can buy.

Facebook, Instagram and TikTok work the other way — they create demand. They interrupt someone scrolling and make them want something they were not actively seeking a moment ago. This is genuinely valuable, but it is earlier in the decision and colder.

For a considered purchase like a week of frozen dinners — where the buyer is weighing trust, price, dietary fit and whether you even deliver to them — captured intent usually converts better than interrupted attention. So the spend order matches the logic of the whole guide:

• Start with Performance Max on Google. It draws on the product feed you already optimised in Chapter 1 and meets buyers at the moment of intent. This is the first rand.
• Add a visual social channel to build demand. Once Google is working, Facebook, Instagram or TikTok are where you create new demand with appetite-appeal video — cheaper reach, but earlier in the decision. Treat this as demand-building, not your conversion engine.
• Let the two jobs run together. Social builds the branded interest; Google captures it when the person finally searches your name or your category. Each makes the other work harder.

What a rand actually buys in South Africa

These figures move with competition, season and your own ad quality, so treat them as orientation, not a quote — but they are useful for sizing a budget and setting expectations.

Platform	Typical SA cost-per-click	Best at	Notes
Google Search / Shopping	~R10–R150+	Capturing intent	Wide range by keyword competition; Performance Max uses your product feed
Facebook / Meta	~R3–R8	Building demand	Among the cheapest Facebook markets globally — well below the world average
Instagram	~R5–R18	Visual demand	Strong for food and lifestyle; runs on Meta’s platform
TikTok	Lowest CPMs of the three	Reach / awareness	Cheap reach, but minimum budgets make it unsuited to micro-testing

Two things to read out of that table.

South Africa is, on the whole, an inexpensive market to advertise in. Meta clicks here run well below the global average, which means a small, well-aimed budget goes further than most merchants expect. The constraint on a frozen brand is rarely the cost of the click — it is whether the product page, the reviews and the checkout (Chapters 1 to 3) are ready to convert the click once it arrives. Cheap traffic to a leaky funnel is still wasted money.

TikTok’s entry cost is real. Its reach can be the cheapest of the three, but it imposes minimum budgets — around a US$20 daily minimum at ad-group level and roughly US$50 at campaign level — which means it is not suited to micro-budget testing. If you are starting with a few hundred rand a month, prove the model on Google and Meta first, and treat TikTok as a step you take once you have budget to feed its algorithm properly.

Spend small, focused, and measured

With the order settled, three disciplines keep a small budget honest:

• Start focused, not broad. Put budget behind your best-selling, highest-margin products first — not the whole catalogue. For a frozen brand, your trial pack (Chapter 2) is often the smartest single product to advertise, because it is the easy yes that opens the subscription relationship.
• Let free and paid run together. Paid listings take the prominent slots; your free listings keep working underneath and in AI answers. You are not choosing between them.
• Measure leads, not clicks. Track which products and queries actually produce orders, not just traffic, and shift budget accordingly. This is the bridge into Chapter 6 — without the numbers, ad spend is guessing with a budget attached.

Where this touches the cold chain

Of every chapter, this is the one where a cold-chain failure costs you the most — because here you have paid for the customer it disappoints.

An ad is a promise made at scale. Every rand you spend driving a customer toward a purchase is a rand that only pays off if the cold chain holds at the other end — because a paid customer whose first delivery arrives thawed is the most expensive kind of bad review you can buy. You did not just lose an organic visitor; you paid to acquire a detractor, and that detractor’s one-star review then suppresses the organic discovery you were also building. Worse, paid acquisition is most efficient when it feeds retention — and a thawed first delivery kills the subscription (Chapter 4) before it starts, turning an acquisition cost into a pure loss.

So the discipline is simple and it is the discipline of the whole guide: do not scale paid acquisition faster than your cold chain can keep its promise. Advertising amplifies your delivery reliability in both directions — it multiplies a brand that delivers well, and it multiplies the damage done by one that does not. (The physical reliability that makes paid acquisition safe to scale is covered in Maintaining the Cold Chain.)

The AI-discovery angle

Paid spend and AI discovery are not separate budgets — the first quietly funds the second.

Paid traffic that converts produces more orders, which produce more reviews and more branded searches — and those are exactly the signals that feed your organic and AI visibility for free. A Performance Max campaign that wins a customer who then leaves a descriptive Google review and tells a friend has bought you something beyond that one sale: it has fed the review text AI engines summarise (Chapter 3) and the branded interest that strengthens organic discovery. The ad budget’s job is to accelerate the free engine, not to become the engine — and well-spent, it makes the free engine spin faster long after the spend stops.

Make your paid products machine-readable: the feed is the ad

There is no new schema to learn for this chapter, and that is the point worth making. Performance Max does not run on bespoke ad creative you build from scratch — it runs on the same product feed you already structured in Chapter 1, and the same Product and Offer schema you have been applying throughout. Your GTINs, your explicit “frozen” attributes, your accurate pricing and availability, your delivery-area data — that structured data is what Google assembles into Shopping ads, Search placements and YouTube cards.

This is why the foundation chapters come first: a clean, frozen-specific feed is not just your free-listings asset, it is your paid asset too. The merchant who skipped the feed work in Chapter 1 cannot run an efficient Performance Max campaign in Chapter 5, because there is nothing good for the machine to assemble. Get the feed right once and it works for free listings, AI answers, and paid — three returns on one piece of structured-data discipline. Verify it, as always, with Google’s Rich Results Test and in Merchant Center’s diagnostics.

The second front: being found by AI assistants

Google still handles most of search, and it is not going anywhere. But AI-assistant referrals are the fastest-growing new discovery channel, and for a considered purchase like a week of frozen dinners, more people now open an assistant and ask it to recommend or compare before they buy. As orientation rather than a quote: by early 2026, Google’s AI Overviews were appearing on roughly a quarter of searches — up from around half that a year before — and brands named inside an AI answer earn materially more clicks than those left out of it. This is the same discovery you have been building for in Chapter 1 and Chapter 3, seen from a second front.

There is now a name for the discipline of earning that mention: Generative Engine Optimisation, or GEO. Where traditional optimisation tries to rank a blue link a customer then clicks, GEO is about being cited inside the answer the assistant gives — named, quoted, recommended. The difference that matters is that it behaves closer to binary than ranking does: on a given question you are either named in the answer or you are absent from it. There is no page two to drift onto.

The reassuring part is that what earns a citation is mostly work this guide has already asked of you. Genuine third-party mentions and earned media are the single largest factor in whether an assistant names you (Chapter 3); clear, answer-first pages that state the thing plainly are the easiest to lift into an answer (Chapter 2); structured data that matches what a human actually sees on the page lets a machine trust it (Chapter 1); and real reviews tell the assistant you are worth recommending. GEO is not a separate discipline learned from scratch — it is the same hygiene as Chapters 1 to 3, done well, plus one thing most merchants never think to check.

The locked door most merchants never check

Here is the part that catches good brands out. An AI assistant does not “know” your shop — when someone asks it a question, it sends a web crawler to read your pages and build its answer from what it finds. If that crawler cannot reach your pages, you cannot be cited, no matter how clean your feed, how sharp your schema or how steady your reviews. And the crawler can be turned away at any of three layers without you ever seeing it happen: a rule in your site’s robots file, a setting in your security or firewall plugin, or a policy at your CDN or hosting edge — very often switched on by a well-meaning “block bad bots” default that was never meant to shut out the assistants your customers now ask. The door looks open from the front; it is quietly bolted at the back. As orientation rather than a quote, audits in early 2026 suggest a meaningful share of online stores — by some counts around one in four — are turning these crawlers away without realising it.

It helps to know that not every crawler does the same job, because the distinction decides what you actually want open. One family reads your page to answer a live question a person has just asked, and can send a real visitor back to you — these are the ones that decide whether you can be found and cited. A second family gathers content to help train the AI models themselves. Allowing the first is what puts you in the running to be named today. Allowing the second is a separate, optional choice about whether your content helps build future models — a legitimate decision either way, and one that does not change whether you get cited. The point is simply not to block the first by accident while you were only ever worried about the second.

The job here is a check, not a build. You are not hand-configuring anything or pasting in code — you are confirming that nothing on your side is quietly turning away the crawlers that earn citations, across both the robots file and the firewall, host or CDN layer. Treat it as auditing the door, in the same spirit as the “know your numbers” discipline of Chapter 6: a thing you verify rather than assume. The quickest path is to ask whoever maintains your site to confirm it, or to bring in a specialist for one focused look — it is a short job for someone who knows where to check.

One note on the horizon, as orientation rather than hype. “Buy inside the assistant” checkout — where an AI agent completes the purchase without the customer ever reaching your site — is real and arriving, but it is rolling out in the United States first, with Canada and Australia expected to follow, and even there it is still settling rather than finished. It is not yet live in South Africa. So the South African merchant’s focus stays exactly where this guide has put it: on discovery and a clean, machine-readable product feed — the foundation that leaves you ready for agentic checkout the day it does arrive, and that earns you citations long before then.

A second horizon note, because it is the question every merchant now asks: as AI moves from recommending to transacting, will it start reading trust the way it reads your product feed? A machine-readable trust layer for AI commerce is indeed forming through 2026 — but it is worth knowing where it is being built, because it is not where most people assume. The verification standards taking shape are set by the payment networks and identity providers, and they run on payment-rail attestations, verified-purchase confirmation and agent-identity checks — not on any single vendor’s “trust score”, and not on your email setup.

What an assistant actually reads when it decides whether to name and recommend you stays exactly what this guide has asked for: clean structured data that matches your page, a verified product feed, and genuine earned reputation (Chapter 3). Email authentication keeps its own important job — being trusted in the customer’s inbox, covered in that chapter — but it is not the rail this layer is built on. The takeaway is the reassuring one: there is no new trust technology to chase here, only the same foundation, kept clean.

Your Accelerate checklist

• Confirm the free foundation (Chapters 1–3) is live before spending a rand.
• Start with Performance Max on Google — capture intent first.
• Add a visual social channel (Facebook, Instagram or TikTok) to build demand, second.
• Budget small and focused — best-sellers and your trial pack first, not the whole catalogue.
• Mind TikTok’s minimum budgets — prove the model on Google and Meta before testing it.
• Measure leads, not clicks — track which products produce orders, and shift budget there.
• Make sure your product feed is clean and frozen-specific — it is your paid asset as well as your free one.
• Confirm no firewall, host or robots rule is turning away the AI assistants’ crawlers — the citation door has to be open, or the rest of this work goes unseen.
• AI-discovery line: paid traffic that converts produces the reviews and branded search that feed your organic and AI visibility — the ad budget accelerates the free engine, it does not replace it.

This is Chapter 5 of the full guide. Previous: Keep Them. Next: Know Your Numbers — the handful of metrics that tell you what’s working. Want the whole thing in one place, plus the consolidated checklists? Get the complete guide.

A note on tone: some of the articles we link to are written in a deliberately blunt, myth-busting register — they challenge the “industry standard” head-on, because the physics demands it. This guide is calmer by design. The engineering underneath is the same.