Decisions · CPG · AI Visibility

General Mills supplied none of the sources AI assistants cited for its categories

We asked ChatGPT, Claude, Gemini and Perplexity the buying questions for eight General Mills categories, from cereal to dog food. Across 94 answers, cheerios.com, bettycrocker.com and the six other brand sites were cited 0 times. That is a machine availability gap, and the company behind it spends $873.6 million a year on marketing.

A brand team at a packaged-food company has a planning line for advertising and a line for distribution. Advertising builds mental availability: the buyer thinks of you when the need comes up. Distribution builds physical availability: the buyer can find you and buy you. Earlier this week we named a third step that now sits between them. Machine availability asks whether the assistant a buyer consults names you, and whose pages it reads before it does.

That post tested a B2B category. This one tests the hardest case for the idea: a portfolio of household brands with decades of advertising behind them. If fame carried into AI answers anywhere, it should carry here. We took eight General Mills brands, wrote down what we expected before pulling any data, and asked four assistants the questions a shopper asks when choosing.

The decision belongs to a brand or portfolio lead, with finance holding the money: does machine availability get its own owner and budget line, or is it assumed to follow from the brand? The two answers lead to opposite plans.

Eight famous brands, zero citations

The Number
0 of 94

AI answers that cited a General Mills brand’s own website, across eight categories: cereal, baking mixes, taco kits, granola bars, dog food, frozen snacks, soup and mac and cheese. No brand’s own site, General Mills’ or a rival’s, appeared among the ten most-cited sources in any category.

Source: CULT+MATH answer-share runs, ChatGPT, Claude, Gemini and Perplexity with web search on, two questions per category asked twice, US, November 19, 2026

Each category got two buying questions: “What is the healthiest breakfast cereal to buy?”, “What is the best boxed cake mix?”, “Which dog food brands do vets recommend?” and so on. Each was asked twice of each assistant. We counted which websites the answers cited as sources. The leading sources held across the two askings in four or five of the top five places in every category, so these are settled answers, not noise.

Decision Map · Who Supplies the Answer
CategoryBrand site citedWho the answers leaned on
Taco kits (Old El Paso)0 of 12tastingtable.com in 8 of 12: one publisher owns it
Soup (Progresso)0 of 12health.yahoo.com in 8 of 12
Dog food (Blue Buffalo)0 of 11dogster.com in 6 of 11, petmd.com in 5
Cake and brownie mix (Betty Crocker)0 of 12foodnetwork.com, allrecipes.com, americastestkitchen.com, 4 each
Cereal (Cheerios)0 of 12health.yahoo.com and healthline.com, 4 each
Frozen snacks (Totino’s)0 of 11eatthis.com in 5, chowhound.com in 4
Mac and cheese (Annie’s)0 of 13Eight sites at 4 each, led by parenting blogs and tasteofhome.com
Granola bars (Nature Valley)0 of 11consumerreports.org in 3; no clear leader

Source: CULT+MATH answer-share runs, November 19, 2026. Counts are answers citing the site, of the answers whose sources could be read. Full tables in the study.

Leading the search did not make the brand the source

Search share is a fair reading of mental availability: people looking for the brand by name. On that measure, Cheerios leads its set at 33.4% of searches against Honey Bunches of Oats, Frosted Flakes and Froot Loops. Annie’s leads its set at 62.7%. Their own sites were cited in 0 of 12 and 0 of 13 AI answers. The brands that win the search did no better than the ones that lose it, because no brand site was cited at all.

Case

Cheerios and the toddler question

November 19, 2026

Asked which breakfast cereal is healthiest and which is best for toddlers, the assistants leaned on health.yahoo.com, healthline.com and a Google redirect, each in 4 of 12 answers, with amazon.com and the BBC behind them. That is ground Cheerios has advertised on for decades. Whatever the assistants concluded, they built it from health writers’ pages, not from cheerios.com.

On a buying question, the source of record is whoever reviews the category, and a brand’s fame does not make its own site that source.

The money side makes the gap sharper. For the fiscal year ending May 31, 2026, General Mills spent $873.6 million on marketing and advertising, 45.2% more than in fiscal 2019. Its gross margin, 33.6%, was the highest among Conagra, Campbell’s and Kraft Heinz, against a median of 28.1%. Its sales fell 5.4%, partly because it sold its US yogurt business to Lactalis in June 2025. The company can pay for attention. On these questions, the attention the buyer finally acts on is assembled from other people’s pages.

We expected the recipe sites to win. They did not.

Before any data came back, we wrote down four explanations and rated them. The one we favored, at roughly even odds, was that General Mills’ recipe-led sites would carry the recipe categories: bettycrocker.com for cake mixes, oldelpaso.com for taco kits. Both were cited 0 times in 12. The explanation we rated unlikely, that third-party publishers carry every category, held in all eight. We also predicted General Mills would lead search in three to five of the eight sets. It led in two of seven.

The limits are real, and they shape what this can claim. The instrument counts the websites an answer cites, not the brands it names, so an answer can recommend Cheerios without citing cheerios.com; this measures the source side of machine availability, not the naming side. Every question was a buying question, and a brand’s own site is not an independent source on whether it is the best; “how do I make…” questions may still go to Betty Crocker. Gemini’s sources come back as a Google redirect, so they could not be read. Every other assistant cited General Mills zero times.

When This Pattern Is Yours
  • Your buyers ask assistants which one to buy — your category is chosen on comparison, not habit.
  • The answers lean on reviewers — asked your buyers’ questions, the assistants cite publishers and lists, not brand sites.
  • The same few publishers repeat — one or two sites carry half the answers, as Tasting Table, Yahoo Health and Dogster did here.
  • Your spend does not reach those publishers — your media plan buys reach, not the reviews the assistant reads.
  • Not yours if your buyers ask how to use the product, not which one to buy. Those questions may well go to your own recipe or help pages; measure them before moving money.
Six Lenses
  • CEO Eight household brands, 0 of 94 citations, while Cheerios and Annie’s lead their search sets. Fame on search is not standing in for machine availability. Name an owner for it in the 2027 plan.
  • CFO $873.6 million in marketing and advertising, and none of it shows up as a source in these answers. Against cutting brand spend on one day’s reading. Hold the budget; fund a quarter of measurement on your own questions.
  • CRO tastingtable.com carried 8 of 12 taco-kit answers and dogster.com 6 of 11 dog-food answers. Against spreading effort across every site. Work with the two or three publishers that own each category’s answers.
  • CPO Recipe sites lost on buying questions, but “how do I make” questions were not asked. Against rebuilding the brand sites now. Test the usage questions before changing the sites.
  • Legal The assistants describe the brands from pages the brands do not control, including health claims about cereal for toddlers. Check what the cited publishers say about your products as you would an ad claim.
  • Risk One day, two questions per category, and one assistant’s sources unreadable. Re-run quarterly and report every figure with its date.

Reconciled. The CEO and CRO want to act now, because the gap is total and the publishers who own the answers are named. The CFO, CPO and Risk want the usage questions and a second reading before money moves. The course that satisfies all six: give machine availability an owner, measure your own buying and usage questions this quarter, and direct partnership and PR effort at the few publishers that already carry your categories’ answers, leaving media budgets where they are until the second reading.

The Path

Ask your three biggest brands’ buying and usage questions this month, and list who the answers cite.

Take two buying questions and two usage questions per brand, worded the way shoppers word them. Ask each twice of ChatGPT, Claude, Gemini and Perplexity with web search on, and record the sites cited. Mark which sites carry half the answers or more, and whether your own site appears on the usage questions.

What it costs
About three hours by hand for three brands, or under a dollar a category through an API. The real cost is holding any shift in media money until the reading is in.
How you’ll know
By the end of the quarter you can name, for each brand, the publishers that own its buying answers and whether your site owns its usage answers. If your site is cited on neither, machine availability is a budget line, not a footnote.

Want this for your portfolio of brands?

We ran it for General Mills with public data. We can run the same measurement for your brands, your categories and the questions your buyers ask, with the predictions registered before the data. A custom brief is $995. The full tables are in the study.

Email Ipalibo to request this Or commission it as a custom brief →
This piece reports how four AI assistants answered given questions on one date and which websites they cited, alongside General Mills’ public filings. It is analysis, not investment or commercial advice, and says nothing about the quality of any product or publisher named. CULT+MATH has no relationship with any organization named in it.