The procurement manager in your category doesn't open Google: they open Word, ask Copilot for the leading suppliers in the sector, and start their assessment from that list. If your name doesn't show up in that AI-generated draft, you won't be added by hand — you simply don't exist in that deal. Being present at the moment the short-lists are formed, before the supplier comparison even begins, is worth more than any sales activity that comes after.
2 billion WhatsApp users are testing Meta AI. Their first questions are “what’s the best X?”. If you’re not in the answer, you’re losing a hidden global channel.
Let me tell you right away why that number opens an article about Microsoft Copilot in Office 365: the principle is identical. There’s a conversational AI interface, a huge pool of users, a search behavior like “who makes the best X?”, and an answer that cites a handful of sources and ignores thousands. Only the usage context changes: on WhatsApp the user asks their phone, inside Office 365 the buyer asks Copilot while preparing a slide for the purchasing committee.
And in there, for a fresh-cut produce company in Lazio — the ones that wash and bag ready-to-eat salads — there’s a slice of revenue at stake without the owner ever realizing it.
What Copilot really does inside Word, PowerPoint and Outlook
Copilot in Office 365 isn’t a generic chatbot pasted next to Word. It’s an assistant that, when you ask it “write me a slide on bagged salad suppliers for retail chains in Central Italy,” does two things at once: it draws on the content of your company tenant (SharePoint files, Outlook emails, OneDrive documents) and queries the web through Bing.
The part that matters to you, the one who sells washed bagged salads from Latina to the supermarket chain, is the second. When Copilot generates a sales presentation, a supplier-scouting email, or a benchmark document for a category manager, it pulls from Bing’s SERPs and decides which companies to name in the draft.
If your brand shows up in those drafts, you’re in the game. If it doesn’t, the buyer won’t add you manually: they’ll narrow the list using the names Copilot has already written for them.
Why this isn’t a new channel, it’s the channel where the buyer builds the short-lists
In the world of research on AI applied to knowledge work, the documented mechanism is that generative models embedded in productivity tools don’t replace search, they precede it. The user doesn’t open Google and then Word: they open Word, ask the assistant, and already find the list of names to validate.
From this follows a practical consequence for your fresh-cut produce company: the retail buyer’s decision moment is no longer only on the public search engine. Part of it has shifted inside the Office 365 interface, where a national chain’s category manager asks Copilot to summarize the supplier landscape of Central-Southern Italy before scheduling the three interviews.
It’s the same principle I explained in the previous articles on tokenization and how AI engines think: the model doesn’t “know” who you are, it recognizes tokens and co-occurrences. If your company name doesn’t co-occur in Bing’s web index with “fresh-cut produce,” “washed salad,” “Agro Pontino,” “retail supplier,” you won’t end up in Copilot’s drafts.
An all-image site, zero structured text.
The reverse engineering: I asked Copilot what it actually cites
I ran a reverse-engineering test to understand what patterns the sources Copilot draws on have. I took 8 typical queries from people who use Office 365 for work — requests like “make a slide on the leading Italian fresh-cut produce makers,” “write me an email to request a quote from bagged salad suppliers in Lazio,” “list 5 farming companies in the Agro Pontino active in fresh-cut produce” — and looked at which brands recurred in the generated drafts.
Across 8 prompts, 14 recurring company names came up. Looking at them closely, 11 of the 14 had at least four of these elements:
- An active Bing Webmaster account and an up-to-date sitemap
- An “about us” page with the address of Latina, Sabaudia, Pontinia or Aprilia written out in full
- A Wikipedia or Wikidata entry for the brand, or for the cooperative/consortium they belong to
- Trade press articles (Corriere Ortofrutticolo, Fresh Plaza Italia, Italiafruit) with the company name + product + geographic area in the same paragraph
- Certifications (BRC, IFS, Global Gap) stated in the homepage text, not just as logos
An indicative test, not a study: 8 queries don’t make research, and Copilot changes its drafts based on the tenant. But the pattern is clear: those who show up have web content formatted as reference material, meaning tables, data, comparisons, certifications in plain text. Those who don’t show up have showcase sites with lots of photos and little extractable text.
Write an “About us” page that names in full Latina, the province, the Agro Pontino, the years in business, the supply chain, the certifications in the text, not just in the logos.
The test you can run this morning in 20 minutes
Before calling the agency, do these three steps yourself. They’re entry level, the real analysis requires professional tools, but they give you a binary sense of where you stand.
Check 1 — Does Bing see you? Open Bing Webmaster Tools, load the domain, see whether the sitemap is indexed. If Bing doesn’t read you, Copilot in Office won’t cite you: the engine under the hood is that one.
Check 2 — Basic search on Bing. Go to bing.com and search `”your brand” “fresh-cut produce” Latina`. If you appear fewer than 3 times in the top 10 results, your brand doesn’t yet have enough local co-occurrence to be picked up by the model.
Check 3 — Ask Copilot directly. If you have a Microsoft 365 subscription with Copilot, try the query “make a slide with the 5 leading makers of washed bagged salads in the Agro Pontino.” If you’re not among the 5, you have your answer about what the retail buyer is seeing when they ask the same question.
Binary threshold: if you fail two of these three checks, the problem isn’t marginal, it’s structural.
The mistakes I see most often in Lazio’s farming companies
When I look at the sites of fresh-cut produce companies between Latina and the Campania border, four patterns come up almost every time.
An all-image site, zero structured text. Beautiful photos of the field at dawn, little readable HTML. Copilot doesn’t cite images, it cites paragraphs.
Vague geographic area. “In the heart of the Agro Pontino” without ever writing Latina, Sabaudia, Pontinia, Borgo San Michele. The model looks for named geographic entities, not evoked ones.
Certifications only as PDF logos. BRC, IFS, Global Gap appear as badges at the bottom of the homepage, but they’re never named in the text. For a model that reads tokens, they don’t exist.
No Wikidata entry. No co-occurrence with the consortium or with other companies in the supply chain in authoritative contexts. This is the point I dig into in the article on author entity recognition and on how to get into the Google Knowledge Graph: the same principle applies to Bing and, by extension, to Copilot.
What to do concretely in the next 30 days
If you sell washed ready-to-eat salads and want Copilot to name you when a national chain’s category manager prepares the supplier slide, these are the moves ordered by return.
- Write an “About us” page that names in full Latina, the province, the Agro Pontino, the years in business, the supply chain, the certifications in the text, not just in the logos.
- Create a “Product spec sheet” page with a table: SKUs, sizes, shelf life, packaging type, certifications, production capacity. This is reference material Copilot loves to pull.
- Register the domain with Bing Webmaster Tools and upload the sitemap. It’s free and it removes doubt number one.
- Check whether a Wikidata entry exists for your consortium or trade association; if you’re a member, make sure your name appears in the cited sources.
- Work with the trade press (Corriere Ortofrutticolo, Fresh Plaza Italia, Myfruit) to get 2-3 articles a year in which the company name and the product co-occur with the geographic area. It’s the kind of signal that triggers the citation proxy in backlinks.
It’s not a magic factor and it isn’t enough on its own, but it moves the needle when the buyer uses Copilot instead of opening the browser.
Visibility in AI answers also runs through your clients’ slides
The thread of this series is always the same: showing up in AI answers doesn’t only mean being first on ChatGPT or Perplexity when an end consumer asks a question. It also means being in the email draft the procurement manager is writing inside Outlook, in the table the category manager is laying out in PowerPoint, in the benchmark document the marketing department of a retail chain is preparing in Word.
Copilot in Office 365 is the piece of this equation that almost none of your competitors is covering, because it looks like “Microsoft stuff” far from sales. It’s exactly the opposite: it is sales, brought forward three days before the phone call.
In the next articles of this series I’ll take you inside the other platforms: Bing Chat in the browser, the differences between consumer and enterprise Copilot, and the relationship between these systems and the underlying Bing engine.
