Google is hiring to generate AI content inside Discover

Google is hiring engineers to generate content for Discover. Three listings on Google’s careers site, read in full on 12 October 2026, describe a Discover Feed Recommendation team that — in Google’s own wording — “leads by recommending AI-generated content (AIGC) to fulfill user interests”. The published United States salary bands for those three roles run from $207,000 to $382,000.

The phrase “AI-generated content” appears in none of the documents Google gives a site owner about Discover. Not in the developer documentation, Last updated 2026-03-09 UTC. Not in the Discover content policies, which carry no update date at all. Both pages were checked twice on 12 October 2026 with an explicit yes-or-no question naming the exact strings, and both reads agreed.

That gap is the whole story, and it is narrower than the headlines around it. A job listing is not a product launch, not a policy and not a measurement. What it is, here, is a dated, publicly readable statement of where a Google team is pointed — written in Google’s recruiting, and nowhere a publisher is told to look.

What the Google job listings say

Three listings were read in full from Google’s careers site on 12 October 2026. None of them carries a posting date, so the date in this article is the date they were read, not the date they appeared. They are documents rather than studies, so no sample size applies to any of them.

The most specific sentence sits in the first of those three. Google writes that the engineer “will build the agentic loops and evaluation frameworks needed to ensure factuality, freshness, coherence, and tone for AI-generated content that empowers users across Google’s proactive surfaces like Discover and Notifications”. Read plainly, that describes an automated editorial pipeline with quality gates, for content Google itself produces and then recommends.

One absence inside the listings is worth recording. Asked explicitly, the Senior Staff listing contains no instance of the word “publisher” anywhere in its body. The word appears once across the three, in the Director+ design role, which asks for experience “building creator/publisher value exchanges” — that is, an exchange presented as something to be built rather than one that exists.

Search Engine Journal, which reported the listings on 11 October 2026, describes a fourth role, a Senior Product Data Scientist on the Discover Experience team (job 84042463571780294), and attributes to it a figure of more than 500 million daily active users for Discover. That listing could not be fetched during this run — the request was refused — so the figure is Search Engine Journal’s rendering and is not quoted here as read from Google.

What Google’s publisher-facing documentation says

Google’s page Discover and your website, Last updated 2026-03-09 UTC, is the document that tells a site owner how Discover works. It states that content is “automatically eligible to appear in Discover if it is indexed by Google and meets Discover’s content policies”, and adds that eligibility is not a guarantee of appearing. Asked on two separate reads whether the strings “AI-generated”, “AIGC”, “AI content” or “machine-generated” appear anywhere in its body, the answer was no to all four, both times.

The Discover content policies page carries no last-updated date. It names ten content policies — dangerous content, deceptive practices, harassing content, hateful content, manipulated media, medical content, sexually explicit content, violent extremist content, violent and gory content, and vulgar language and profanity — plus three feature-specific policies covering ads and sponsored content, misleading content and transparency. Checked in the same yes-or-no form, it contains no “AI-generated”, no “AI”, no “automatically generated” and no “scaled”.

Google’s two announcement channels say nothing either. The Search Central documentation changelog, read live on 12 October 2026, carries no entry dated 9, 10, 11 or 12 October; its newest is 8 October, “Added documentation on the UGC Fresh Data Program”, preceded by 1 October, “Updated guidance on using generative AI content”. The Search Central blog feed’s newest post is 6 October 2026, announcing curated YouTube playlists. For completeness: the Search Status Dashboard was read with a cache-busting query string and a month-scoped prompt and carries no October 2026 incident of any kind, the September 2026 spam update having closed at 2026-10-08T08:00:00+00:00.

One source this article wanted and could not get: Google’s Using generative AI content guide, which the changelog shows was edited on 1 October 2026, returned a client error on three separate attempts during this run. Its wording is therefore not quoted here, and no claim in this article rests on it.

What it changes in practice

Nothing in a site’s Discover eligibility changed today. The documented rule is the same rule it was in March: be indexed, meet the content policies. No ranking system is known to have moved, and how Google actually gets from crawling to serving is unaffected by who Google is interviewing.

What changes is the thing a Discover strategy is implicitly betting on. Discover traffic has always been a bet that Google will send readers to an article because the article is the best available answer to an interest. A feed team staffed to generate and quality-control its own content is a second supply of answers inside the same surface, competing for the same slots. That is a different bet, and today is the first dated Google artefact that describes it in Google’s own words.

It also sharpens a measurement problem the analytics level of the course keeps running into. Search Console reports Discover performance, but it reports impressions and clicks for your URLs. It does not report what else was in the feed beside them, so a decline caused by more AI-generated items competing for slots looks identical in your data to a decline caused by your content getting worse.

What to do this week

• Export your Discover report now, before anything changes. Take sixteen months of impressions and clicks out of Search Console and store it with today’s date. If the feed’s composition changes over the next two quarters, that export is the only before you will have.
• Separate Discover from Search in your reporting. They are different surfaces with different volatility, and averaging them hides exactly the movement worth watching.
• Record the two documents and their stamps. Discover and your website at 2026-03-09 UTC and the undated content policies page. If either gains a paragraph about generated content, that is the day the rules changed — not today.
• Do not rewrite anything in response to a job listing. There is no documented eligibility change to optimise for, and there is no published threshold to clear.
• Watch both changelogs, not one. Google’s documentation is split between the Search Central changelog and the separate crawling-infrastructure changelog, and an edit logged in one does not appear in the other.

Where the industry genuinely disagrees

Three questions this story touches are contested, and this article does not settle any of them.

Is a job listing evidence of a product direction? One camp holds that recruiting is the most honest thing a company publishes — a team cannot hire for work it does not intend to do, salary bands are real money committed, and waiting for an announcement means reporting the news after everyone has already adapted. The other holds that a listing is a wish, that teams are staffed for projects that are cancelled, that no listing carries a posting date or a launch date, and that a page which can be edited or closed without notice cannot be re-checked by a reader later. Nobody can settle it, because Google publishes no mapping from a role it advertises to a feature it ships.

Where does scaled content abuse begin? Google’s policy targets content produced at scale primarily to manipulate rankings, explicitly regardless of how it was produced, and publishes no threshold. One camp reads Google generating content for its own surface as beside the point: the policy governs what is submitted to Search, not what Google builds inside Discover, and a platform producing its own feed items is not competing for rankings. The other reads it as the clearest available illustration that the rule is about purpose rather than production, and that the industry has been arguing about the wrong variable. No measurement separates them, because no public threshold exists to measure against. Our AI for SEO level presents this as an open question rather than a settled one.

Should a publisher keep investing in a surface whose operator is becoming a producer on it? One camp says yes without hesitation: Discover is large, the traffic is real today, and withdrawing from a channel because of a hiring page forfeits revenue on a guess. The other says a channel where the platform supplies competing inventory cannot be forecast, and that the time is better spent on surfaces where supply is not controlled by the party ranking it. Nobody has published a comparison, because the generated inventory is not measurable from outside.

The author’s opinion — Txema Hermoso

I am defending one thing here and it is documentary: on 12 October 2026, Google’s recruiting describes AI-generated content as the Discover feed team’s product, and Google’s publisher-facing documentation for Discover does not contain the phrase. Both halves are reproducible by anyone with a browser.

I am not saying Google has launched AI-generated articles in Discover, that your traffic is about to fall, or that this is the end of Discover as a referral channel. Google published no measurement, named no surface, promised no date and changed no policy, and I will not invent a mechanism to fill that gap. The listings tell you where a team is pointed. They do not tell you when, how much, or at whose expense.

The one judgement I will offer is about the word that is missing. Three listings describe building quality gates for content Google generates, and the only mention of publishers across them is a design role asked to build a “creator/publisher value exchange” — a thing described as not yet existing. If you are planning a strategy for AI-driven surfaces, that absence is worth more planning time than the salary figures.

What is still unknown

Whether any of these roles has been filled. Whether the work ships, and to which Discover markets. What kind of content is meant — summaries of existing articles, original items, or something closer to the AI previews Discover has already tested. Whether generated items will be labelled. Whether they will displace publisher links or sit beside them. Whether Search Console will ever report feed composition. And whether Google’s Discover documentation will be updated before any of this reaches users, or after.

Sources

• Google Careers, “Staff ML Engineer, Search Discover Feed Recommendation” (job 126262492455150278), careers listing, read 12 October 2026. No posting date shown.
• Google Careers, “Senior Staff ML Software Engineer, AI Generated Content Quality, Search Discover Feed” (job 111538254716510918), careers listing, read 12 October 2026.
• Google Careers, “Principal, UX, Search Discover” (job 82901196271952582), careers listing, read 12 October 2026.
• Google, Discover and your website, Last updated 2026-03-09 UTC, read twice 12 October 2026.
• Google, Discover content policies, no last-updated date shown, read 12 October 2026.
• Google, Search Central documentation changelog, read 12 October 2026. Newest entry 8 October 2026.
• Google, Search Central blog feed, read 12 October 2026. Newest post 6 October 2026.
• Google, Search Status Dashboard, incidents read 12 October 2026 with a cache-busting query string. No October 2026 incident.
• Matt G. Southern, Search Engine Journal, “Google Is Hiring To Generate AI Content For Discover”, 11 October 2026.

Every figure in this article is a published salary band or a job identifier taken from the listing itself. None of the sources here is a study, so none of them has a sample size.