How FMCG brands grow retailer sales with AI content infrastructure

An FMCG brand does not own the shelf it sells on. It rents space on somebody else's.

That has always been true in physical retail. What has changed is that the digital shelf now has three separate gatekeepers, and every one of them reads content before a human ever does. Google and AI assistants decide whether your brand appears in an answer. The retailer's own search bar decides whether your product appears in a result set. Your own brand site decides how much either system trusts you in the first place.

Most FMCG brands are structurally weak on all three. Not because the marketing is weak, but because the content was never written for this job.

TL;DR

• FMCG content competes on three surfaces at once: AI and search answers, the retailer's internal search bar, and the brand's own website. Most brands optimise for none of them.

• The copy is the wrong kind of copy. FMCG product content is typically written as specification, not as benefit or occasion, and it is thin. Shoppers do not search in specifications.

• The difficulty is combinatorial, not editorial. Brands multiplied by SKUs multiplied by markets multiplied by retailers. A mid-size portfolio can require over 70,000 distinct content variants before a single seasonal refresh.

• Winning content is richer, seasonal and trend-reactive. The same product sells against completely different queries in December and in July.

• The brand website is the biggest untapped signal. Brands that do not sell direct tend to run thin, unlocalised sites, which is precisely the source search engines use to understand the brand.

• It works when it is solved as infrastructure. A world leader in personal care increased e-commerce revenue by 12% with Newtone, on a deliberately limited set of products and markets.

FMCG brands do not own the shelf they sell on

A direct-to-consumer retailer controls its own product pages, its own search, its own merchandising. An FMCG brand controls almost none of that. The product page on the retailer's site is built from a feed the brand supplies. The search results are ranked by the retailer's engine. The AI answer that recommends a product is assembled from whatever the models can find and trust.

This means FMCG content has to perform in three different environments, each with different rules.

The retailer search bar is the one most brands underestimate. It is not a semantic engine. It matches the words in your feed against the words the shopper typed. If your description says "keratin complex, 250ml" and the shopper types "repair damaged coloured hair", you are not in the result set. You did not lose on price or ranking. You were never eligible.

The content FMCG brands have is the wrong content

FMCG product copy is usually produced by, or for, the people who make the product. It reads accordingly. It describes what the thing is, in the vocabulary of the people who formulated it, rather than what it does for the person buying it.

The result is content that is simultaneously too technical and too thin. Technical, because it leads with composition and format. Thin, because once the specification is listed there is nothing else on the page.

This is the same attribute depth problem we measured across retail catalogues in Why 60% of your product catalog is invisible to AI search. In FMCG it is more acute, because the product itself is simple and the differentiation lives entirely in the content.

The real problem is combinatorial, not editorial

Every FMCG marketer already knows their content should be richer. That is not the blocker. The blocker is that "richer" has to be delivered across four multiplying dimensions at once.

Multi-brand portfolio. Large catalogue. Many markets, each with its own language and its own regulatory constraints on claims. Many retailers, each with its own feed specification, character limits, attribute taxonomy and keyword priorities.

Here is what that multiplication looks like for a mid-size enterprise FMCG portfolio. These are illustrative figures, not research findings, but the structure is what matters.

Now apply a realistic production cost. Generic AI tooling plateaus at roughly 70% of publishable quality, and closing the remaining 30% takes around 23 minutes of human editing per asset once review and correction are included.

144,000 assets at 23 minutes each is 55,200 hours. That is roughly 31 people doing nothing else, all year, to stand still.

This is the same structural failure we documented for retail in The content execution gap costing retailers their AI search rankings. FMCG simply has more multiplying dimensions than most retail categories, so the gap opens faster.

Seasonality: the same product, two completely different searches

FMCG demand is seasonal in a way that most catalogue content completely ignores. The product does not change. The reason someone buys it changes entirely.

A concrete example from recent work. A fresh goat cheese. In December, the relevant shopper intent is stuffing for a Christmas turkey: a filling, a festive recipe, a quantity per bird. In July, the same SKU is a barbecue product: cheese on toast, grilling, an aperitif with friends.

Two entirely different vocabularies. Two different sets of queries. Two different sets of occasions, recipes and pairings. One unchanged product page that mentions none of it, because it was written once, at launch, and describes a soft cheese made from goat's milk in a 150g tub.

Multiply that by the 72,000 variants above and seasonality stops being a nice-to-have and becomes the single largest missed revenue opportunity in the catalogue. Freshness also compounds the problem on the AI side, where recency is a ranking input, as we set out in Your category pages have an expiry date.

Trends move faster than content calendars

Seasonality is at least predictable. Trends are not.

FMCG is the category where consumer trends emerge and peak fastest: a protein claim, a viral recipe format, a new dietary preference, an ingredient that becomes suddenly desirable or suddenly suspect. The window between a trend becoming searchable and becoming saturated is measured in weeks.

A content operation that takes eight to twelve days per brief, then routes through legal and brand approval, then waits for the next feed sync, cannot participate in that window. By the time the content ships, the trend has moved. The brands that capture trend demand are not the ones with better trend-spotting. They are the ones that can act on a trend across thousands of SKUs and a dozen markets within the same week.

The bottleneck is the process, not the writing

It is worth being precise about where FMCG content actually gets stuck, because it is rarely at the drafting stage.

Every one of these is a governance problem. That is exactly why generic AI writing tools fail here: they solve drafting, which was never the constraint, and they cannot enforce brand voice, market regulation or retailer specification. This is the distinction between a tool and infrastructure that we set out in Why retail content teams are moving from AI tools to AI infrastructure.

Localisation deserves particular attention in FMCG, because food, personal care and household categories are where claim regulation is strictest and where shopper vocabulary is least translatable. We covered why literal translation destroys commercial performance in How brand-quality translation drives international revenue.

The brand website is the biggest untapped signal

Here is the blind spot that costs FMCG brands the most, and the one that is cheapest to fix.

Because most FMCG brands do not sell direct, their own website is treated as a brand-image asset rather than a commercial one. It is often a handful of pages, a brand story, a product range shown as images, and frequently a single language for a business that sells in twenty markets.

Search engines and AI models do not treat that site as decoration. They treat it as the authoritative source on the brand. It is where they resolve what the brand is, what it sells, which products belong to it, what claims it makes, and whether it is a credible entity at all. Every retailer listing is a secondary, derivative source. The brand site is the primary one.

The leverage here is unusually high. A brand site is a controlled environment: no retailer feed specification, no character limits, no syndication delay. It is the one surface where an FMCG brand can publish the rich, original, non-commodity content that AI systems reward, exactly as we argued in Google just settled the GEO debate.

Full catalogue coverage, localised properly, with occasion and usage content, recipes, routines and FAQ depth, turns a brochure site into the strongest discoverability asset the brand owns.

What this looks like when it works

A world leader in personal care applied exactly this approach with Newtone: richer benefit-led and occasion-led content, adapted per market and per retailer, refreshed against seasonality, and deployed with brand voice and regulatory rules enforced at generation rather than at review.

It was deliberately scoped. A limited set of products. A limited set of markets. A controlled test rather than a full portfolio rollout.

The revenue number is the headline, but the operational change is what makes it repeatable. Once content production is infrastructure rather than a project, extending from a limited scope to the full portfolio is a configuration change, not a new budget cycle.

A practical sequence for FMCG content teams

If you run content, e-commerce or digital for an FMCG portfolio, the order of operations matters more than the ambition.

1. Fix the brand site first. It is the primary entity signal, it is fully under your control, and it has no retailer constraints. Full catalogue, all selling markets, native localisation.

2. Audit shopper vocabulary per retailer. Compare the words in your feed against the words shoppers type into each retailer's search bar. The gap is usually the fastest commercial win available.

3. Rewrite from specification to benefit and occasion. Lead with the outcome, then add usage, pairing and proof. Target 15 to 20 enriched attributes rather than 6 to 8.

4. Build a seasonal calendar per category, not per campaign. Decide in advance which SKUs change intent across the year, and what the content becomes in each window.

5. Move compliance upstream. Encode brand voice, market claim rules and retailer specification into generation, so review becomes verification rather than rewriting.

6. Start scoped, then scale. Prove the lift on a defined set of products and markets before committing the full portfolio.

Frequently asked questions

What is AI content infrastructure for FMCG?

It is a system that generates, governs and syndicates product and brand content at catalogue scale, with brand voice, market regulation and retailer specification enforced at the point of generation. It differs from an AI writing tool in that it handles the governance and distribution layers, not only the drafting.

Why does FMCG product content underperform on retailer sites?

Because it is written as specification rather than as benefit or occasion, and retailer search engines match literally against the words in the supplied feed. If the shopper's vocabulary does not appear in the content, the product is not eligible to be returned, regardless of its ranking or price.

How often should FMCG content be refreshed?

At minimum, in line with the seasonal intent shifts of each category, which for most FMCG products means at least twice a year. Trend-sensitive categories need a faster cadence. Recency is also a ranking input for AI systems, so static content decays in visibility even when it remains accurate.

Does a brand website matter if the brand does not sell direct?

Yes, and arguably more. Search engines and AI models use the brand site as the authoritative source for understanding the brand entity, its catalogue and its claims. Retailer listings are treated as secondary sources. A thin or single-language brand site weakens visibility across every retailer where the brand sells.

How do you keep AI-generated content compliant across markets?

By encoding the rules before generation rather than checking after it. Brand guidelines, per-market claim regulation and per-retailer specification are applied as constraints on the model, so the output is compliant by construction and human review becomes verification rather than rewriting.

Where Newtone fits

Newtone is content engineering infrastructure for retail and FMCG brands. We do not sell an AI writing tool. We build the systems that let brands produce on-brand, non-commodity content across their full catalogue, in every market and for every retail partner, at the quality and cadence the digital shelf now requires.

Trained on each brand's editorial guidelines, claim rules, SEO priorities and existing content, our models generate publish-ready product content, category and brand-site pages, occasion and recipe content, FAQs and localised assets, with the brand voice consistency and attribute depth that search engines and AI systems reward. This is the approach we set out in Generative engine optimization, a new visibility frontier for enterprise retail.

For FMCG specifically, the value is in the multiplication. The same governed system produces the winter version and the summer version, the French claim set and the German one, the retailer A feed and the retailer B feed, without a proportional increase in headcount, agency spend or approval time.