GEO tools: what they measure and what you can check yourself

Generative engine optimisation means being named and cited in answers from assistants. That includes ChatGPT with search, Perplexity, Google AI Overviews and AI Mode, Claude, Gemini and Microsoft Copilot.

Tools track three things. Mentions: your brand or product name appears in the prose. Citations: your page is linked or footnoted as a source. Share of answer: how much of the answer is about you, often scored by turns, lines or bullets.

They score this across a set of buyer prompts. For example, “best email warmup tools for startups”, “how to price A/B testing SaaS”, or “yourproduct vs rivalname”. They run these prompts over time and across assistants, then show trends and gaps.

Search backed assistants fetch from a web index. They cite pages that answer a question directly, with clear structure, consistent names, lists, comparisons and quotable figures.

• OpenAI: GPTBot for training, OAI-SearchBot for ChatGPT search indexing, ChatGPT-User for on demand fetches.
• Anthropic: ClaudeBot for training and model improvement, Claude-SearchBot for index, Claude-User for on demand. Older tokens like anthropic-ai and Claude-Web still appear.
• Google: Googlebot for Search and pages cited in AI Overviews, Google-Extended to control Gemini training and grounding. Google-Extended does not change Search or AI Overviews.
• Perplexity: PerplexityBot for index, Perplexity-User on demand.
• Microsoft: Bingbot for Bing Search and Microsoft Copilot.
• Apple: Applebot for Siri and Spotlight, Applebot-Extended to control Apple Intelligence training.
• Others seen: CCBot, Bytespider, meta-externalagent and meta-externalfetcher, Amazonbot, cohere-ai, DuckAssistBot.

Check that your robots.txt allows the answering bots you want. Then check your logs or analytics for fetches by these tokens. If they never reach you, you cannot earn citations, even with great content.

Sampling matters. If a tool asks broad prompts like “email marketing tools”, results skew to huge brands. If it asks buyer prompts like “best email warmup tools for LinkedIn outreach”, more niche products surface.

• Seed set: they start from keyword lists, your sitemap and rival pages.
• Clustering: they group similar prompts to avoid double counting.
• Assistant mix: they query ChatGPT with search, Perplexity, Google AI Overviews or AI Mode, Claude, Gemini and Microsoft Copilot.
• Cadence: they refresh on a schedule, typically daily or weekly, then roll up 28 days to smooth noise.
• Location and intent: some prompts get different answers by country or intent. Tools that do not note this can mislead.

Ask any vendor how they pick prompts, how often they run them, and how they detect citations versus plain mentions. If they cannot explain it in one page, be careful with the output. You can still use your own prompt list to cross check their graphs.

You can run a simple check by hand. Do this every 28 days. Keep it the same each month so you can compare.

1. 1

   Make a prompt list

   Write a repeatable list of buyer prompts you can cover in an hour. Mix head and long tail. Examples: “best A/B testing tools for Shopify”, “yourproduct vs rivalname”, “pricing page examples for SaaS”, “alternatives to rivalname for freelancers”. Keep them stable month to month.

2. 2

   Pick the assistants

   Use ChatGPT with search, Perplexity, Google AI Overviews or AI Mode, Claude, Gemini and Microsoft Copilot. If an assistant has multiple modes, use the one that cites sources.

3. 3

   Ask and capture

   Run each prompt once in each assistant. Copy the full answer into your sheet. Save the cited URLs and the brands named. Note the order of mentions.

4. 4

   Score share of answer

   Example scoring: for each prompt and assistant, score 1 point if you are cited, 1 point if named, and 1 extra point if you are the first named. Do the same for two key rivals. This gives a quick index.

5. 5

   Tally by page and topic

   Group the citations by your URL. For example, if /pricing wins citations, note it. If guides/getting-started is never cited, note that too. Map prompts to site sections.

6. 6

   Log what the assistant quotes

   Copy the phrases, figures and bullets it lifts. Note if it quotes a list, a table, a comparison, or a stat. These are the shapes you must provide.

A simple sheet has columns for date, assistant, prompt, named first, named later, cited URL, rival cited URLs, and notes. It typically takes under an hour after the first month once you have a template.

Assistants cite pages with clear, extractable answers. They prefer pages with stable names for products and people. They lift bullets and tables readily. They like explicit comparisons and crisp figures they can quote. Examples below are illustrative, not measured.

• Name your product the same way on every page, including /pricing, /features and /guides/getting-started.
• Use a summary at the top. One paragraph with who it is for, what it does, and one figure or limit, like “free plan, 3 seats” (example).
• Use lists and comparison tables. For example, a table on /pricing that compares plans by seats, usage, and support hours.
• Answer the exact question on the page. A query like “is yourproduct HIPAA compliant” needs a section with that heading and a one line answer before the detail.
• Quote your own figures. Example: “Average setup time: 12 minutes” with a short method note. Assistants pull numbers with a label and unit.
• Write a plain /alternatives/yourproduct page and a /comparison/yourproduct-vs-rival page. Assistants cite these when users ask versus and alternatives prompts.

A fixed page looks like this: the h1 is the exact task, the first paragraph answers it in one sentence, then a short list of facts, for example three to seven bullets. Yourproduct.com/guides/ai-overview-examples could open with a one line definition, then three examples, each with a cited source link. That is easy to quote and to cite.

Look at what the answer cites. You will usually see roundups, comparisons and feature pages. Your task is to earn a place on those lists and to match their page shapes on your own site.

• Match the cited roundup pages
• Email the editors of the specific roundups the assistant cites. Offer a short, factual blurb. If they update lists monthly, ask for the next cycle.
• Ship your own comparison pages
• Publish /comparison/yourproduct-vs-rival and /alternatives/rival. Use neutral headings and a fact table. Link them from /features and /pricing.
• Tighten product naming
• Unify the product name, company name and plan names across the site. Assistants miss you if the names drift.
• Add missing facts and shapes
• If answers quote max limits, add them. If they show tables, add a table. If they lift bullets, use short bullets.
• Fix crawl access
• Confirm that OAI-SearchBot, Claude-SearchBot, PerplexityBot, Bingbot and Googlebot are allowed. Check with a live robots.txt tester and server logs.
• Be present where answers shop
• Find the most cited roundups and directories for your topic. Submit your product. If they charge or have a form, follow it. If they only add by editorial choice, send a short fact sheet.

Check back on the same prompts next month. You should see your name appear in mentions before citations rise. Keep the foot on the facts and the shapes assistants can lift cleanly.

Access is a precondition. If the answering crawlers cannot fetch, you will not be cited. Blocking training is your call, but do not block search and answering bots if you want visibility in assistants as of 2026.

1. 1

   Allow the answering and search bots

   In robots.txt, allow Googlebot, Bingbot, OAI-SearchBot, Claude-SearchBot and PerplexityBot. If you block by default, add explicit allows.

2. 2

   Decide on training controls

   Use Google-Extended, Applebot-Extended and any vendor controls you choose for training. These do not affect Search ranking or AI Overviews, but they can limit use in model training.

3. 3

   Verify with a tester and logs

   Use a robots.txt tester to confirm access. Then check server logs for hits from the expected tokens. If you never see them, check IPs and any CDN blocks.

4. 4

   Make sources clear on page

   Link to your own evidence and documentation. Use stable URLs. Assistants pull footnotes from clear, canonical URLs.

If you rely on JavaScript rendering for core content, ensure the raw HTML still exposes the summary, the lists and the tables. Many crawlers index the rendered page, but you do not control when or how deep they render. Fallback HTML makes the quotable bits safe.

You do not need a dashboard. A sheet works. Keep one tab per month so you can compare changes side by side without lag or sampling drift. The example values below are illustrative, not measured.

At the end of the tab, sum points by assistant and prompt. Graph total points over 28 days. Keep a short narrative under the graph: what changed, what you shipped, which pages gained citations. This will steer your next month’s fixes.

Buy a tool when your manual list gets too long, or you need alerts. You still need to define prompts, check robots access, and ship the fixes. Tools help you see more prompts and more assistants faster. They do not rewrite your pages or place you on cited lists for you.

If you try one, test it against your sheet. Check that its prompts match your buyers. Check that its citation detection matches what you see in the answers. Keep your manual run once a month as a control. It keeps you honest when the charts move.