Stop Guessing GEO Keywords: Bing Grounding Queries

Quick Takeaways (read this first)

• Grounding queries are not user prompts—they’re the retrieval searches Bing’s AI runs to find sources to cite. That makes them the most actionable “AI-retrieval keyword” dataset you can get today.
• Your unit of optimization is the cited passage, not the page. Add “citation-ready” blocks (80–140 words) that match common qualifiers like vs, best, template, pricing, steps, pitfalls.
• Use Topics + Intents to avoid random edits: pick one Topic cluster and one intent (comparison/how-to/pricing), then refactor 1–3 pages to match the AI’s retrieval language.
• Measure with Compare + Citation Share (relative presence per grounding query), not just “total citations.” That’s how you prove week-to-week progress.
• You can run this as a 7-day sprint: export → cluster → choose 10–20 grounding queries → write blocks → ship → compare → iterate.

What’s changed: Bing quietly gave you the AI’s “search terms”

If you’ve been running GEO/AEO efforts off made-up prompts (“What would a user ask ChatGPT?”), you’ve felt the pain: the tracker says you’re “visible,” but nothing is repeatable. The missing link has been retrieval language—the exact phrasing an answer engine uses when it goes looking for sources.

Bing Webmaster Tools now exposes that missing layer through AI Performance reporting, including grounding queries—the internal searches Bing’s AI uses to retrieve web pages before it cites them. Microsoft documents this directly in the AI Performance report help doc. In other words: this isn’t what your users typed, and it isn’t what you guessed an LLM might be prompted with. It’s what the AI actually searched to find citable passages.

Over the last few weeks, this has become operational—not just interesting—because Microsoft added workflow-grade features like Intents, Topics, Citation Share, and Compare (preview) and a Topics Portfolio (beta) that clusters grounding queries into themes. See Microsoft’s announcements on Intents/Topics/Citation Share/Compare and Topics Portfolio (beta).

Grounding queries vs. SEO keywords vs. “prompt tracking” (why this matters)

Most teams are mixing three different “keyword” concepts as if they’re interchangeable. They’re not.

1) Traditional SEO keywords (human queries)

These are what people type into search engines. They’re great for demand modeling and traffic forecasting, but they don’t tell you how an AI rewrites or qualifies a query when it’s hunting for sources.

2) GEO trackers based on invented prompts

Prompt-based visibility tools can be useful for directional monitoring, but they’re easy to game accidentally. You can “win” a prompt by matching its wording while still being absent from the engine’s real retrieval layer.

3) Grounding queries (AI retrieval searches)

Grounding queries are the AI’s own reformulations—often adding qualifiers like:

• Comparison: “X vs Y,” “alternatives,” “pros and cons”
• Selection: “best,” “top,” “recommended,” “for small business”
• Implementation: “steps,” “how to,” “template,” “checklist”
• Commercial validation: “pricing,” “cost,” “free,” “trial,” “eligibility”
• Trust/constraints: “requirements,” “limitations,” “compliance,” “policy”

This is why grounding queries are uniquely actionable: they tell you how the engine looks for evidence. If you match that language with extractable passages, you become easier to cite.

A contrarian (but practical) insight: your “GEO keyword list” should be AI rewrites, not head terms

In classic SEO, you might start with a head term (“employee onboarding software”) and expand into long-tail variants. In GEO/AEO, we’ve found the better starting point is the rewrite the engine uses to retrieve sources:

• “employee onboarding software pricing”
• “employee onboarding software vs HRIS”
• “employee onboarding checklist template”
• “how to separate onboarding by role best practice”

Those qualifiers are not fluff. They’re often the difference between: (a) a page that ranks but never gets cited and (b) a page that becomes the AI’s evidence.

The 7-day playbook: turn grounding queries into a measurable GEO workflow

This sprint is designed for in-house SEO/AEO leads who need something they can do every week, report on, and iterate.

Day 1: Export grounding queries and pick one Topic cluster

1. Open Bing Webmaster Tools → AI Performance.
2. Filter to the last 28 days (or 7 days if volume is high enough).
3. Export Grounding Queries along with pages and citation metrics available in your view.
4. If you have it, use Topics / Topics Portfolio to cluster queries. (This is exactly what Topics Portfolio beta is meant to do: group grounding queries into themes so you can build a backlog instead of a spreadsheet graveyard.)

Tip: Don’t pick a cluster because it’s biggest. Pick it because you can ship improvements in 7 days (existing pages, clear intent, manageable scope).

Day 2: Choose 10–20 grounding queries with “citation opportunity”

Your goal is not “more keywords.” Your goal is a short list where you can plausibly win citation share quickly.

Use this selection rubric:

• High citations, low satisfaction on your site: grounding queries that cite you sometimes, but your page doesn’t explicitly answer the qualifier.
• High-intent qualifiers you don’t cover: “pricing,” “requirements,” “template,” “steps,” “mistakes,” “vs.”
• One intent at a time: pick either comparison or how-to or pricing/eligibility for the sprint.

This is where Bing’s Intents and Compare views matter. Microsoft positions these as AI visibility insights to help you understand not just that you’re cited, but where you stand relative to others for a given AI scenario (details here).

Day 3: Map grounding queries to 1–3 target pages (and decide: refactor vs. create)

For each grounding query, choose the “best existing page” that should win the citation. Then decide:

• Refactor if the page already covers the topic but doesn’t match the qualifier language (common).
• Create only if the qualifier represents a genuinely distinct need (e.g., “pricing” deserves a dedicated page or section with tables and update timestamps).

Common mistake: creating 20 new pages for 20 grounding queries. That’s classic SEO thinking. In AI citation land, you usually win faster by making one page extremely extractable for a cluster of closely related qualifiers.

Day 4: Rewrite for retrieval language (H2s, intros, and definition blocks)

Now you make your pages look like what the AI is searching for. This is not keyword stuffing; it’s aligning headings and answer blocks with the engine’s retrieval phrasing.

A simple page pattern we recommend

1. Qualifier-matching H2s: include the key rewrite in the heading (naturally).
2. 80–140 word citation-ready block immediately under the H2.
3. Evidence payload: bullets, steps, constraints, examples, and a short table if relevant.
4. Internal anchor links so the AI (and users) can jump to the exact section.

Good vs. bad example (comparison intent)

Bad (hard to cite): “Our platform is flexible and powerful for modern teams…” (no comparison frame, no crisp claims)

Good (citable): “Onboarding software vs. HRIS: Onboarding tools focus on task orchestration (checklists, role-based steps, reminders), while an HRIS is the system of record for employee data (payroll, benefits, compliance). Choose onboarding software when you need cross-team workflows; choose an HRIS when you need data governance.”

Day 5: Add “citation-ready blocks” for each qualifier (the real unlock)

If you only do one thing from this post: do this.

Grounding queries are heavily qualifier-driven. So instead of writing one long narrative, create extractable answer blocks the AI can safely quote.

Template: 80–140 word citation-ready block

• Sentence 1: direct definition/answer (no throat-clearing)
• Sentence 2: constraints or “when to use”
• Sentence 3: steps/checklist OR pitfalls OR requirements
• Sentence 4: brief example or decision rule

Real-world examples (3) of grounding-query-driven blocks

These examples show how grounding queries often look (qualifiers included) and how you’d respond.

Example 1: “pricing” qualifier

Grounding query: “employee onboarding software pricing per user”

Block to add under an H2 like “Employee onboarding software pricing (what to expect)”:

Employee onboarding software pricing is typically charged per active employee or per admin seat, with entry plans often starting around a low monthly minimum and scaling by headcount and features (integrations, e-sign, automation). To compare vendors fairly, ask for (1) the billing unit (employee vs seat), (2) minimum contract length, (3) included integrations, and (4) overage rules for seasonal hiring. If your team hires in cohorts, look for tiered pricing or “active month” billing to avoid paying for dormant accounts.

Actionable tip: add a small table with “Billing unit / Typical add-ons / Questions to ask” and a “Last updated” date.

Example 2: “template” qualifier

Grounding query: “customer onboarding checklist template B2B SaaS”

Block under “Customer onboarding checklist template (B2B SaaS)”:

A B2B SaaS customer onboarding checklist template should cover: kickoff (success criteria + stakeholders), technical setup (SSO, roles, integrations), first value milestone (complete one core workflow), enablement (admin training + end-user materials), and adoption tracking (usage events + QBR schedule). Keep the checklist role-based—admin steps differ from end-user steps—and assign an owner and due date per item. The fastest way to improve time-to-value is to define one “activation event” (e.g., first automation runs) and build the checklist backwards from that outcome.

Actionable tip: offer the checklist as both an on-page table and a downloadable CSV/Notion copy.

Example 3: “how-to + pitfalls” qualifier

Grounding query: “how to separate onboarding flows by role common mistakes”

Block under “How to separate onboarding by role (and avoid mistakes)”:

To separate onboarding flows by role, start by listing the first 30-day outcomes for each role, then map tasks to owners (manager, IT, HR, buddy). Build one shared “company baseline” flow (policies, security, benefits) and layer role modules (tools access, training, certifications). Common mistakes include duplicating tasks across flows (creates conflict), not gating access requests by start date (creates security risk), and skipping a single source of truth for task status. A practical rule: if a task’s owner differs by role, it belongs in a role module.

Actionable tip: add a short checklist titled “Role-based onboarding QA” to improve extractability.

Day 6: Validate “extractability” (format + anchors + on-page evidence)

Citation eligibility is partly about trust, but it’s also about mechanics: can the system confidently extract a clean answer?

Before you ship, run this checklist:

• Answer appears above the fold for the section (don’t bury it after a 600-word story).
• Use one idea per paragraph (AI systems prefer clean, unambiguous spans).
• Add section anchors (e.g., “#pricing”, “#template”, “#mistakes”).
• Support claims with specifics: numbers, constraints, definitions, and steps.
• Remove “marketing fog” near the answer block (superlatives without evidence reduce quotability).

If you’ve been seeing “wins that vanish” in live answer engines, it’s often because your page entered the candidate set but didn’t remain the best extractable citation. That’s exactly the problem we addressed in our sprint write-up on candidate sets and verifiability: LIVE‑GEO Proves GEO Copy Tweaks Fail: 7‑Day Fix.

Day 7: Measure the right KPI—Compare + Citation Share per grounding query

Total citation counts are noisy. They go up and down with seasonality, model changes, and surface changes. What you can control—and report weekly—is whether you’re winning a specific micro-space.

Use Bing’s Compare and Citation Share to:

1. Pick 1–3 grounding queries you optimized for.
2. Record your baseline Citation Share (or whatever comparative metric is available in your account).
3. Re-check after shipping changes (give it time to be crawled and reflected).
4. Iterate: if you’re cited but not dominant, add one more citation-ready block that matches a close qualifier.

If you want a deeper KPI framing specifically for Bing’s AI reporting, we broke down why Citation Share is the first “real” GEO KPI in BWT here: Bing’s Citation Share: The First Real GEO KPI in BWT.

How to turn grounding queries into content briefs (without bloating your site)

Once you have a Topic cluster, you’ll be tempted to create a new page for every query. Don’t. Instead, create a brief that turns a cluster into a section architecture.

Brief template (copy/paste)

• Topic cluster: [Name from Topics Portfolio or your label]
• Primary intent: comparison | how-to | pricing/eligibility | troubleshooting
• Target page(s): 1–3 URLs
• Grounding queries (10–20): list + tag each with qualifier type (vs/best/template/pricing/steps/mistakes)
• New H2s to add: 5–8 headings that mirror qualifiers
• Citation-ready blocks: 1 block per H2, 80–140 words
• Evidence assets: table, checklist, short example, definitions
• Measurement plan: Compare + Citation Share for 3 priority grounding queries

Common mistakes we see (and how to fix them fast)

Mistake 1: Treating grounding queries like normal SEO keywords

If you stuff them into title tags or write a “keyword list” blog post, you miss the point. Grounding queries tell you how to structure answers so the AI can retrieve and cite.

Fix: convert each qualifier into an H2 + citation-ready block.

Mistake 2: Optimizing the whole page instead of the cited passage

AEO/GEO is passage-first. The AI often extracts 2–5 sentences.

Fix: write 80–140 word blocks with a definition + constraints + steps/pitfalls + example.

Mistake 3: Reporting “citations” without a denominator

“We got 30 citations” isn’t a strategy. It’s a screenshot.

Fix: report Citation Share movement for 3 grounding queries per week.

Mistake 4: Ignoring the policy reality: AI is becoming inseparable from search

If your org still treats AI answers as optional, the market is moving the other way. Recent reporting shows Google’s stance that schools opting out of AI features may lose access to Search—an aggressive signal that AI and search distribution are being fused at the platform level (see AP News coverage and TechRadar’s analysis).

Fix: make GEO/AEO a weekly operating cadence, not a quarterly experiment.

FAQ: Grounding queries (the questions your team will ask)

Are grounding queries the same as what users type into Copilot?

No. Grounding queries are the AI’s retrieval searches—internal reformulations used to find sources to cite. They may be derived from user intent, but they’re not the raw prompt.

How many grounding queries should we optimize for at once?

Per 7-day sprint: 10–20 grounding queries inside one Topic cluster, mapped to 1–3 pages. That’s usually enough to see movement without creating chaos.

Do we need to create new pages for every qualifier (pricing, template, vs)?

Usually no. Start by adding qualifier sections to your strongest relevant page. Create a new page only when the qualifier implies a distinct job-to-be-done (e.g., pricing pages, compliance requirements, or downloadable templates).

What’s a realistic success metric in 7 days?

Don’t promise “doubling citations.” Aim for: (1) improved extractability (new blocks shipped), (2) better coverage of one intent, and (3) measurable movement in Compare/Citation Share for 1–3 grounding queries once recrawled and reflected.

What to do next (action steps you can assign today)

1. Pull your last-28-days grounding queries from Bing AI Performance and export them.
2. Select one Topic cluster (Topics Portfolio if available) and pick one intent.
3. Choose 10–20 grounding queries with qualifier patterns you can address quickly.
4. Refactor 1–3 pages: add qualifier-matching H2s + 80–140 word citation-ready blocks.
5. Track 3 queries using Compare + Citation Share; iterate with one additional block per underperforming qualifier.

Try this workflow inside aeotool.ai

If you want to turn this into a repeatable weekly operating system, we built our dashboard to help you manage AEO/GEO work like a sprint: pick a topic, map it to pages, track changes, and report outcomes.

You can sign up and explore the AEO tool dashboard here: https://aeotool.ai/register. And if you want lightweight, page-by-page checks while you edit, install our Chrome extension: AEO Analyzer Chrome extension.