---
title: "Does It Fix Anything? A 20-Point Checklist for AI Overviews Tools"
description: "Use 20 demo questions to tell whether an AI Overviews vendor watches, recommends or executes—and find the labour its subscription leaves with your team."
url: "https://groas.com/post/does-it-fix-anything-or-just-watch-the-2"
image: "https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/8d39dde9-e4ae-43ef-b11c-b7e4f61b2b9e.png"
published: "2026-10-14T05:47:51.555Z"
modified: "2026-10-14T05:47:51.628Z"
---

October 14, 2026 · 9 min read

# Does It Fix Anything? A 20-Point Checklist for AI Overviews Tools

[Alexander PerelmanHead Of Product @ groas](https://groas.com/author/alexander-perelman)[LinkedIn](https://www.linkedin.com/in/alexander-433793253/)

![At dusk, three contractors face a broken shopfront: one watches through binoculars, one points with a clipboard, one is up a ladder fixing it.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/8d39dde9-e4ae-43ef-b11c-b7e4f61b2b9e.png)

In this article

1. [A. Find the gaps, not just a visibility score](#a-find-the-gaps-not-just-a-visibility-score)
2. [B. Change the page instead of exporting homework](#b-change-the-page-instead-of-exporting-homework)
3. [C. Fix the layer that can block the answer](#c-fix-the-layer-that-can-block-the-answer)
4. [D. Prove what changed, including what failed](#d-prove-what-changed-including-what-failed)
5. [E. Price the work the subscription leaves behind](#e-price-the-work-the-subscription-leaves-behind)
6. [Score the vendor, then check who owns the work](#score-the-vendor-then-check-who-owns-the-work)

Most tools sold to get you cited in AI Overviews will never touch your site. I have sat through enough demos to know the pattern: a share-of-voice chart, prompts where a competitor gets named, then an export button that puts your team in charge of the fix. Use these 20 questions to sort a vendor into **watcher, adviser or doer** before you pay. Score each answer **0** if the vendor cannot show it, **1** if it flags or drafts work your team must finish, and **2** if it handles the relevant work and shows proof. For monitoring questions, that proof is prompt-level evidence; for price questions, it is written terms. For fixes, it means a live change you can audit. Ask to see each answer in the product, not on a slide.

## A. Find the gaps, not just a visibility score

![Ink cartoon of a dashboard watching a leak while unfinished pages pile up](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/a8af9d54-86bd-471f-9923-e9d523fcd61c.png)

1. **Track prompts separately by engine.** Ask to see the same buyer questions in AI Overviews, ChatGPT, Perplexity and Claude, rather than one blended score. A plausible demo answer is: “Here are your 50 prompts, by engine, with their run dates.” Check the cap before you score it: [Profound Starter tracks ChatGPT only](https://ecommercefastlane.com/ai-visibility-tools-for-ecommerce/), while [Peec Starter offers 50 prompts across three selected models](https://ecommercefastlane.com/ai-visibility-tools-for-ecommerce/). A fourth engine is not covered just because its logo appears in the deck.
2. **Show the buyer questions you miss, by engine.** Ask for the absent price, comparison and product-fit prompts, not merely an overall share-of-voice percentage. “You are absent on 14 of 22 price and comparison prompts in AI Overviews” gives you a place to start. A single visibility number does not tell you which page to change or which question it needs to answer. Score 0 if the vendor cannot get from its chart to the individual prompt.
3. **Name who gets cited instead, with the winning URL.** Ask for a table that pairs each missed prompt with the cited competitor and page. “Competitor X appeared” is weaker than a URL you can inspect for its answer, structure and relevance. The practical test is the one used in this [tool price check](https://www.youtube.com/watch?v=b6PElzj2cjE): who gets named instead of you? If the tool cannot show that, its gap report stops one step too early.
4. **Explain a gap in terms you can act on.** Ask what differs between your page and the cited one: perhaps the competitor has a comparison table or FAQ markup where you do not. Treat that as a proposed fix to test, not a guarantee of a citation. Many trackers identify the winner and hand the diagnosis back to you. That is the [analytics trap](https://www.mersel.ai/blog/geo-beyond-analytics-to-execution): a report can be accurate and still leave every useful decision undone.

## B. Change the page instead of exporting homework

5. **Draft or edit the answer on the relevant page.** Ask to see the proposed answer block, comparison table or FAQ against the URL that lost the prompt. A generic content brief deserves a 1, not a 2: someone still has to turn it into copy that fits the page. The demonstration should connect one missed buyer question to a specific edit, so you can judge whether the tool understood the gap or merely generated more words.
6. **Publish the approved change to the live site.** Ask the vendor to show the handoff from draft to CMS and then the changed URL. An export button, CSV or copy-paste instruction leaves publishing with you, however polished the draft looks. Few tools [publish without handing the keyboard back](https://growganic.io/blog/ai-visibility-tools-compared), which makes this the fastest way to separate an adviser from a doer. Score 2 only when the vendor handles the live change, with your approval where needed.
7. **Keep publishing after the first audit.** Ask what happens to missing buyer questions next month: who plans, creates and publishes the next comparison or FAQ page? A one-time list of suggested topics is not a cadence. Some execution-focused services describe [ongoing GEO content production](https://www.mersel.ai/blog/geo-beyond-analytics-to-execution); ask any vendor to show what it has shipped, rather than accepting a monthly target in a proposal. Score 1 if your team must keep the schedule moving.
8. **Add structure an engine can lift.** Ask for the before-and-after page, including a direct answer, a useful table where appropriate, and relevant FAQ or HowTo markup. [Structured HTML and schema](https://www.mersel.ai/blog/geo-beyond-analytics-to-execution) are part of the work, not decorative extras in a content brief. A tool that rewrites body copy but cannot show the resulting page and markup has not finished the job. Score the shipped structure, not the promise to add it.

## C. Fix the layer that can block the answer

9. **Check crawler access by bot, not by brand name alone.** Ask which search crawlers can reach the page and which training crawlers are blocked. [GPTBot and OAI-SearchBot serve different purposes, as do ClaudeBot and Claude-SearchBot; Googlebot and Google-Extended are not interchangeable](https://mediaofficers.com/insights/ai-crawler-optimization-robots-txt-cloudflare-headers). A blanket “we allow AI” is not a useful answer. The vendor should show the relevant rules and say who changes them if search access is blocked.
10. **Distinguish verified bot fetches from spoofed requests.** Ask how the crawl report confirms that a request came from the bot it names. A User-Agent string alone can be spoofed; the check should use the relevant published verification information for [OpenAI, Anthropic, Perplexity and Google bots](https://mediaofficers.com/insights/ai-crawler-optimization-robots-txt-cloudflare-headers). If the vendor sells crawl insights but cannot show its method or the underlying log, do not treat its bot counts as proof that your pages were reached.
11. **Ship rendering, schema and internal-link fixes.** Ask for a before-and-after example: a page that rendered poorly, lacked relevant markup or sat orphaned, and the live change that addressed it. [Structured HTML, markup and crawlable rendering](https://mediaofficers.com/insights/ai-crawler-optimization-robots-txt-cloudflare-headers) matter before a good answer can be useful. Score 1 if the tool opens a developer ticket; score 2 if the vendor gets the fix shipped. For a closer look at this work, see [how to fix technical SEO issues that hurt AI visibility](https://groas.com/post/fix-technical-seo-issues-hurt-ai-visibility).
12. **Check snippet controls and WAF blocks.** Ask whether security rules stop relevant crawlers and whether page-level snippet settings restrict what can appear in search. A good answer points to the affected rule or template, identifies who owns the change and verifies the result. Teams can [block a bot or restrict a snippet](https://mediaofficers.com/insights/ai-crawler-optimization-robots-txt-cloudflare-headers) while wondering why a page is hard to surface. A watcher reports the absence. A doer helps remove the block.

## D. Prove what changed, including what failed

13. **Keep a dated action log with a reason for each change.** Ask for an entry that names the URL, the edit and the missed prompt that motivated it: “March 4: rewrote the pricing answer block after losing six comparison prompts.” A monthly impressions PDF is not an action log. Without dates and reasons, you cannot tell whether the vendor responded to a gap, changed something unrelated or simply reported a fluctuation after the fact.
14. **Track citations before and after each edit, by prompt.** Ask to see the prompt, the edit date and subsequent citation checks in one view. This does not prove that one edit alone caused a citation to move, but it lets you inspect the sequence instead of trusting a victory slide. The [tracker-versus-doer distinction](https://groas.com/post/tracker-vs-seo-suite-vs-done-for-you-whi) is simple here: one charts visibility; the other can connect its work to the result it monitors.
15. **Report the pages that did not gain citations.** Ask for the unchanged prompts alongside the wins, plus the next proposed edit. “Three pages shipped; one gained citations and two did not” is more useful than a green arrow covering the whole account. If every report is up and to the right, you cannot see what to stop doing. Score 2 when failures remain visible and the vendor owns the next step, not just the celebration.

## E. Price the work the subscription leaves behind

16. **Get prompt and engine caps in writing.** Ask for a limits table before you sign. Published examples show why: [Otterly Lite lists $29 for 15 prompts across four engines](https://cituna.com/compare); [Peec Starter lists $95 for 50 prompts across three models](https://cituna.com/compare); and [Profound Starter has been listed at $99 for ChatGPT, with broader coverage starting at $399](https://ecommercefastlane.com/ai-visibility-tools-for-ecommerce/). Those figures are comparison points, not a substitute for a current quote. Score 0 if the vendor will not state the limits.
17. **Price extra engines and prompts before you need them.** Ask what it costs to add the next model or expand the prompt set, and get the answer in writing. Add-ons and higher tiers change the bill: the [tool comparison](https://ecommercefastlane.com/ai-visibility-tools-for-ecommerce/) lists engine-dependent upgrades, while this [price check](https://www.youtube.com/watch?v=b6PElzj2cjE) lists Semrush AI Toolkit at $99 per domain and Ahrefs Brand Radar from $199, with custom prompts extra. The lesson is not to choose by the lowest entry price. Price the coverage you intend to use.
18. **Define what a tracked prompt and a refresh consume.** Ask whether one question run on two engines counts once or twice, how often it refreshes and what happens to unused capacity. Those terms decide how much of your buyer-question list fits in the plan. A trial with [50 prompts across three engines and a fixed prompt set](https://cituna.com/compare) may let you inspect the interface without testing your own coverage. Score the limits you can actually use, not the headline count.
19. **Put your team’s hours beside the subscription fee.** Ask who writes, reviews, publishes and handles developer tickets, then estimate those hours at your loaded rate. The [monitoring-only cost comparison](https://www.mersel.ai/blog/geo-beyond-analytics-to-execution) describes plans at $300 to $3,000 a month alongside 20 to 40 hours of internal work. That is the hidden line item. A $29 tracker is cheap only if the work it creates already has an owner with time to do it.
20. **Name who maintains the fix after month one.** Ask who notices a slipped citation, revisits the page, publishes the next edit and checks the prompt again. “Your team” is a legitimate answer, but it makes the subscription a monitoring expense, not an execution service. Get the owner and expected internal hours in the quote. This is the question demos tend to skip because it reveals whether the vendor’s job ends when your team’s begins.

## Score the vendor, then check who owns the work

Add the 20 scores for a maximum of 40. Use the bands as a **labour test**, not a substitute for watching a demonstration:

- **0–13: watcher.** It identifies prompts and competitors. You supply the writing, publishing, technical fixes and maintenance. Buy it if you have people ready to do that work; budget their hours as well as the fee.
- **14–27: adviser.** It drafts or diagnoses some fixes, but your team still gets them live and follows up. Ask who handles approvals and developer tickets before you call the saved drafting time a saving overall.
- **28–40: doer.** It identifies a gap, gets the change live, records the action and monitors the prompt afterward. Do not award this label on points alone: items **6, 11, 14 and 20** must each earn a 2. Ask for one dated, prompt-level example of a shipped edit and its subsequent citation checks.

That is why I put groas ahead of a monitoring-only tool: its model pairs autonomous search execution and action logs with a human strategist who owns direction, rather than stopping at a chart. It still has to pass the same live-change and follow-up questions. Use this [list of tools that help you get featured, not just monitored](https://groas.com/post/tools-to-get-featured-in-google-ai-overviews) to compare vendors, then make each one show the work.

**Ask item 20 last:** when a citation slips in three months, who rewrites the page? If the answer is your team, the cost is not just the subscription. It is the writing, publishing, developer time and maintenance the demo left off the bill.

## Frequently Asked Questions

### Should an AI visibility tool show my prompts separately for each AI engine?

Yes. Ask to see the same buyer questions in AI Overviews, ChatGPT, Perplexity and Claude, rather than one blended score. Check the engine cap before you sign, because some starter plans cover only one engine or a limited prompt set, and a fourth engine is not covered just because its logo appears in the vendor's deck.

### Why isn't an overall visibility score enough to know what to fix?

A single visibility number does not tell you which page to change or which question it needs to answer. Ask for the individual missed prompts, such as price and comparison questions, plus a table pairing each missed prompt with the cited competitor's URL so you can inspect that page's answer, structure and relevance.

### How can I tell if an AI Overviews tool actually makes changes or just exports a to-do list?

Ask to see the handoff from draft to CMS and then the changed URL. An export button, CSV or copy-paste instruction leaves publishing with your team, however polished the draft looks. Only score full marks when the vendor handles the live change itself, with your approval where needed.

### What page structure does an AI visibility fix need to include?

The vendor should show a before-and-after page including a direct answer, a useful table where appropriate, and relevant FAQ or HowTo markup. Structured HTML and schema are part of the work, not decorative extras in a content brief. A tool that rewrites body copy but cannot show the resulting page and markup has not finished the job.

### How do I check whether AI crawlers can actually reach my pages?

Ask which search crawlers can reach the page and which training crawlers are blocked, by bot: GPTBot and OAI-SearchBot serve different purposes, as do ClaudeBot and Claude-SearchBot. A User-Agent string alone can be spoofed, so a crawl report should use published verification information from OpenAI, Anthropic, Perplexity and Google to confirm the request came from the bot it names.

### How should a vendor prove its changes actually led to citations?

Ask for a dated action log that names the URL, the edit and the missed prompt that motivated it, plus a view showing the prompt, the edit date and subsequent citation checks together. This does not prove one edit alone caused a citation to move, but it lets you inspect the sequence. Also ask for the pages that did not gain citations, alongside the wins.

### What hidden costs should I look for in an AI visibility subscription?

Get prompt and engine caps in writing, price extra engines and prompts before you need them, and define what a tracked prompt and a refresh consume. Then put your team's hours beside the fee: monitoring-only plans are described at 300 to 3,000 dollars a month alongside 20 to 40 hours of internal work. Ask who rewrites the page when a citation slips in three months.

## Pay For Results, Not For Hours

Businesses buy the outcome, agencies resell it, and groas answers for it either way.

[See If You Qualify](https://groas.typeform.com/to/xC1bQNUT)

## Structured data

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://graph.groas.com/entity/groas","name":"groas","url":"https://groas.com/","logo":{"@type":"ImageObject","url":"https://groas.com/icon-512.png","width":512,"height":512},"sameAs":["https://www.linkedin.com/company/groas/"]},{"@type":"WebSite","@id":"https://groas.com/#website","name":"groas","url":"https://groas.com/","inLanguage":"en-US","publisher":{"@id":"https://graph.groas.com/entity/groas"}},{"@type":"WebPage","@id":"https://groas.com/post/does-it-fix-anything-or-just-watch-the-2#webpage","url":"https://groas.com/post/does-it-fix-anything-or-just-watch-the-2","name":"Does It Fix Anything? A 20-Point Checklist for AI Overviews Tools","description":"Use 20 demo questions to tell whether an AI Overviews vendor watches, recommends or executes—and find the labour its subscription leaves with your team.","inLanguage":"en-US","isPartOf":{"@id":"https://groas.com/#website"},"about":{"@id":"https://graph.groas.com/entity/groas"},"primaryImageOfPage":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/8d39dde9-e4ae-43ef-b11c-b7e4f61b2b9e.png"},"breadcrumb":{"@id":"https://groas.com/post/does-it-fix-anything-or-just-watch-the-2#breadcrumb"},"dateModified":"2026-10-14T05:47:51.628Z","mainEntity":{"@id":"https://groas.com/post/does-it-fix-anything-or-just-watch-the-2#article"}},{"@type":"BreadcrumbList","@id":"https://groas.com/post/does-it-fix-anything-or-just-watch-the-2#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://groas.com/"},{"@type":"ListItem","position":2,"name":"Blog","item":"https://groas.com/blog"},{"@type":"ListItem","position":3,"name":"Does It Fix Anything? A 20-Point Checklist for AI Overviews Tools","item":"https://groas.com/post/does-it-fix-anything-or-just-watch-the-2"}]},{"@type":"BlogPosting","@id":"https://groas.com/post/does-it-fix-anything-or-just-watch-the-2#article","url":"https://groas.com/post/does-it-fix-anything-or-just-watch-the-2","mainEntityOfPage":{"@id":"https://groas.com/post/does-it-fix-anything-or-just-watch-the-2#webpage"},"isPartOf":{"@id":"https://groas.com/blog#blog"},"headline":"Does It Fix Anything? A 20-Point Checklist for AI Overviews Tools","description":"Use 20 demo questions to tell whether an AI Overviews vendor watches, recommends or executes—and find the labour its subscription leaves with your team.","datePublished":"2026-10-14T05:47:51.555Z","dateModified":"2026-10-14T05:47:51.628Z","image":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/8d39dde9-e4ae-43ef-b11c-b7e4f61b2b9e.png"},"author":{"@id":"https://groas.com/author/alexander-perelman#person"},"publisher":{"@id":"https://graph.groas.com/entity/groas"},"keywords":"ask, vendor, answer, prompts, score, fix, overviews, page","wordCount":1940,"inLanguage":"en-US"},{"@type":"Person","@id":"https://groas.com/author/alexander-perelman#person","name":"Alexander Perelman","jobTitle":"Head Of Product @ groas","description":"Ex Goldman Sachs and Ex Stanford Computer Science","image":"https://groas.com/media/blog/4f17cac3a81acc224cc1b2cfab8f94bbc295a3089d13e8421125771ef8d4064f.jpg","worksFor":{"@id":"https://graph.groas.com/entity/groas"},"sameAs":["https://www.linkedin.com/in/alexander-433793253/"]},{"@type":"FAQPage","@id":"https://groas.com/post/does-it-fix-anything-or-just-watch-the-2#faq","mainEntity":[{"@type":"Question","name":"Should an AI visibility tool show my prompts separately for each AI engine?","acceptedAnswer":{"@type":"Answer","text":"Yes. Ask to see the same buyer questions in AI Overviews, ChatGPT, Perplexity and Claude, rather than one blended score. Check the engine cap before you sign, because some starter plans cover only one engine or a limited prompt set, and a fourth engine is not covered just because its logo appears in the vendor's deck."}},{"@type":"Question","name":"Why isn't an overall visibility score enough to know what to fix?","acceptedAnswer":{"@type":"Answer","text":"A single visibility number does not tell you which page to change or which question it needs to answer. Ask for the individual missed prompts, such as price and comparison questions, plus a table pairing each missed prompt with the cited competitor's URL so you can inspect that page's answer, structure and relevance."}},{"@type":"Question","name":"How can I tell if an AI Overviews tool actually makes changes or just exports a to-do list?","acceptedAnswer":{"@type":"Answer","text":"Ask to see the handoff from draft to CMS and then the changed URL. An export button, CSV or copy-paste instruction leaves publishing with your team, however polished the draft looks. Only score full marks when the vendor handles the live change itself, with your approval where needed."}},{"@type":"Question","name":"What page structure does an AI visibility fix need to include?","acceptedAnswer":{"@type":"Answer","text":"The vendor should show a before-and-after page including a direct answer, a useful table where appropriate, and relevant FAQ or HowTo markup. Structured HTML and schema are part of the work, not decorative extras in a content brief. A tool that rewrites body copy but cannot show the resulting page and markup has not finished the job."}},{"@type":"Question","name":"How do I check whether AI crawlers can actually reach my pages?","acceptedAnswer":{"@type":"Answer","text":"Ask which search crawlers can reach the page and which training crawlers are blocked, by bot: GPTBot and OAI-SearchBot serve different purposes, as do ClaudeBot and Claude-SearchBot. A User-Agent string alone can be spoofed, so a crawl report should use published verification information from OpenAI, Anthropic, Perplexity and Google to confirm the request came from the bot it names."}},{"@type":"Question","name":"How should a vendor prove its changes actually led to citations?","acceptedAnswer":{"@type":"Answer","text":"Ask for a dated action log that names the URL, the edit and the missed prompt that motivated it, plus a view showing the prompt, the edit date and subsequent citation checks together. This does not prove one edit alone caused a citation to move, but it lets you inspect the sequence. Also ask for the pages that did not gain citations, alongside the wins."}},{"@type":"Question","name":"What hidden costs should I look for in an AI visibility subscription?","acceptedAnswer":{"@type":"Answer","text":"Get prompt and engine caps in writing, price extra engines and prompts before you need them, and define what a tracked prompt and a refresh consume. Then put your team's hours beside the fee: monitoring-only plans are described at 300 to 3,000 dollars a month alongside 20 to 40 hours of internal work. Ask who rewrites the page when a citation slips in three months."}}]}]}
```