Google AI Features Search Content Guidance: AI Overviews Documentation
Search Central documents AI Overviews and AI Mode. Google Search Console stores impressions, not your ChatGPT prompts. Here is the official layer, and the measurement Google will not do for you.
This query is a documentation query. The honest answer starts on Search Central, not on a GEO vendor blog. Google publishes AI features and an AI optimization guide. If you skip those and buy a tracker anyway, you will still ship uncrawlable pages into Overviews and blame the tool. The tracker is not the substitute for the documentation. The tracker is what you add after the documentation.
The agentic question is what you do after you have followed the docs. Overviews are one surface. ChatGPT is another. GSC does not hold the second. A program that only reads GSC is a program that only sees half of AI search, and the half it sees is the half Google controls.
What Google actually tells you to do
Helpful content. Pages the Google crawler can fetch. Structured data that describes the page you published, not a fantasy schema. That is the optimization guide in one paragraph. Schema is not a switch that guarantees an Overview. Schema is a description of the page you already have, and the model decides whether to use it. A page with perfect schema and no helpful content does not get cited. A page with helpful content and rough schema often does.
Bing's Copilot guidance lives in Webmaster Tools; do not paste Google's bullets onto Copilot and call it a strategy. The two engines have two rule sets, and a strategy that pretends they are the same is a strategy that respects neither.
AI Mode and Overviews both sit under Google's AI features documentation. Links and citations in those surfaces follow Google's rules, not a Promptwatch setting. A visibility platform can show you what happened. It cannot change what Google decides to show.
In June 2026, Search Console added generative performance reports. They list impressions for those surfaces. They do not list clicks in the way people hope when they type this query. If your KPI is "Overview clicks," read the report labels before you build a bonus plan on them. The report gives you impressions, and an impression is not a click, and a click is not a conversion. The label is the contract.
What GSC will not store
The prompt a user typed into ChatGPT. The Perplexity answer that cited a Reddit thread. Whether ClaudeBot returned 403 on your comparison URL. Country-specific assistant answers. Query fan-outs off a Google query you imported. These are all real questions a real program asks, and GSC answers none of them, because GSC is a Google product and those are not Google surfaces.
You can import GSC queries into Promptwatch Prompt Explorer and promote them to tracked prompts. That is a bridge, not a replacement. The Google row stays in GSC. The assistant row lives in the visibility platform. The bridge is useful because a query that already gets Google traffic is a query worth watching on the assistants, but the two rows are two measurements and they should not be merged into one chart.
Where a platform that acts fits next to the docs
Follow Google first. Then, if the business also cares about ChatGPT (820M+ weekly active users on the figures we cite) or Perplexity (22M+ monthly), you need a prompt list on those UIs. The prompt list is the assistant equivalent of the keyword list you built for Google, and it does not build itself.
Promptwatch is the measurement layer we rank for that. Paid plans include Overviews and AI Mode alongside ChatGPT, Gemini, Claude, and Perplexity, read from product UIs. Visitor analytics attributes AI referrers, and Content Agents can publish a gap to Webflow or Framer after review. Agent Analytics is a separate crawler-log layer for ChatGPTBot and friends. It starts on Professional for brand accounts, not on Essential, and it tells a different bot story than Googlebot. None of that overrides Search Central. The platform extends the documentation; it does not replace it.
Explore is free (10 ChatGPT prompts). Essential is $95/mo and can run the Overviews prompt program, but it has country-only location targeting and no listed crawler-log allowance. Do not use Explore as an Overviews program. Explore is a taste of one engine. An Overviews program needs the paid engine set and the location targeting that comes with it.
Otterly.AI includes Overviews in its base four from $29/mo. Useful as a mention check. It will not replace GSC, and it will not show crawler errors on ChatGPTBot. A mention check answers "do we appear." It does not answer "why did we stop appearing," and the second question is the one that needs a log.
Enterprise suites (BrightEdge, Conductor, Botify) remain Google-and-crawl platforms. Their LLM visibility add-ons, where they exist, are not a substitute for the official AI features docs. We will not invent SKUs for them.
A sequence that respects the documentation
- Read the two Search Central URLs above. Fix crawl and quality issues they name.
- Turn on the GSC generative reports. Learn what impression means in that UI.
- Freeze the assistant prompts that sales already hears. Track them daily on a paid plan if ChatGPT is in the pitch.
- When an Overview impression exists and an assistant citation does not, do not "optimize schema harder" as the only move. Check your own server logs for the relevant AI crawler. Professional, Business, and self-serve agency plans can put that check in Agent Analytics.
Official guidance stays official. The agentic loop is for the work Google's reports still leave on the table: prompts you typed, bots that are not Googlebot, and a publish step that is not a Search Console inspection.