An AI citation is rented, and the lease is 11 days

1. Re-earn the citation on a schedule, and on every engine

Profound published the number on Tuesday. It tracked 883,000 pages (opens in a new tab) across seven answer engines for a year. The median cited page fell to half its peak share in 11 days. 78% did so inside two weeks. Profound sells the tracker this was measured with, so read the shape, not the decimals.

Gauge supplied the other half the day before. Across more than 500 million bot events (opens in a new tab), about 600,000 requests carried the ChatGPT-User agent. None asked for llms.txt. Gauge sells agent analytics and states no period or site count.

So the file you shipped for discovery is not read at the discovery step, and the citation you earned there is gone in a fortnight. Both are a flow.

Am I affected? If your plan for AI answers is a file and one content sprint, yes. What replaces it? Cadence and spread. Correlation between engines was 0.03 to 0.09, and a page cited by five or more of the seven was 5.5 times as likely to hold half its peak after eight weeks.

There is a third surface. Profound also watched Meta’s shopping agent (opens in a new tab) on retail sites. Through the open browser it returned 4 products from 2 merchants. Through Meta’s product catalog, about 60 from 30 sites (50 prompts, one category, “small, exploratory” by its authors’ account). The agent reads the feed. Squarespace drew the same conclusion for domains with an open MCP server (opens in a new tab): no key, no account, two read-only tools, purchase left to the checkout link.

The trade-off: a feed or an open tool hands your catalog and prices to anyone, competitors included.

Your move:

• List the pages that carry your citations today. Put a refresh date on each and republish their facts at least monthly.
• Run the free check on all seven engines, not one. Protect the pages that show up on five.
• Expose the lookup step as data: a price feed or a read-only MCP tool with no auth. Keep the purchase behind the account.

Takeaway: a citation is a lease, not a deed. Renew it on a calendar, on every engine, and give the agent a feed to read.

2. Publish the meter for the builder and the plan for the team

DigitalOcean priced the same compute twice in two days. Thursday’s pricing page (opens in a new tab) listed the raw meters: Harness Runtime from $0.0535 an hour, an Action Gateway from $0.10 per 1,000 calls, inference from $0.05 per million tokens. Friday’s Agent Droplets (opens in a new tab) wrapped the same meters in plans: $5 of credit with no card, $50 a month at 15% off usage, $200 at 20% off. SSO, MFA and audit logs at every tier, and “no seat charges”.

Metronome’s CEO explained why on Stripe’s blog (opens in a new tab) the same day. Token billing belongs in your cost model, not on the invoice. The progression is tokens, then unified credits, then a price on the output. One in six Stripe users past key revenue milestones were rolling out hybrid pricing in August (Stripe’s own book; Metronome sells the billing layer).

The meter is what the builder tests you on. The plan is what the team budgets. The trade-off is margin: the discount is the product.

Your move:

• Keep the raw meter on the page. An agent or a builder reads it first.
• Wrap it in a flat plan, with the discount stated as a percentage off that meter so the two pages reconcile.
• Price the security features at zero seats, and say so in the plan’s first line.

Takeaway: one price list, two readers. The meter earns the trial, the plan earns the budget.

Quick hits

• GitHub shipped the audit, not just the count. (opens in a new tab)
  Its security agent found 24 real bugs in named Android apps, and the post publishes the taskflows and the command to rerun them on your repo, with the cost stated.
• CodeRabbit sells the budget, not the review. (opens in a new tab)
  Its OpenAI Marketplace pitch is to apply “your existing OpenAI spend commitment” through the OpenAI account team, so the listing is a procurement shortcut, and reads best written as one.
• Neon resized its free tier for agents, not people. (opens in a new tab)
  Storage per project doubled to 1 GB with the 100-project cap unchanged, for agents creating “lots of projects, most of them idle”, so the tier is sized for a reader that creates dozens.
• Reddit switches off RSS on 13 November. (opens in a new tab)
  The public API closes for good by March 2027, so subreddit monitoring run from outside breaks this quarter, and Reddit’s own Devvit is the only path left.

Open questions

1. Does Argon get a dated price by Oct 31?
   Google’s announcement (opens in a new tab) gives $2/$10 introductory and $4/$20 after with no end date, and the API pricing page (opens in a new tab) does not list the model: a date there makes it a term, silence leaves a price nobody can budget.
2. Does a third vendor ship a claim-later tier by Oct 31?
   Railway and Neon are still the only two, and WorkOS’s agent registration (opens in a new tab) docs name no adopters: a third auth.md with an anonymous method, or an accountless provision with published caps, makes it the agent free tier.