SolCrys Logo

AEO Fundamentals

How to turn an AI visibility report into a work plan

An AI visibility report can list dozens of gaps and still leave a team with nothing to start on Monday. Here is how to get from that list to shipped, remeasured work.

Updated

Questions this guide answers

  • How do I turn an AI visibility report into a work plan?
  • How should I prioritize AI visibility gaps?
  • What should I fix first when my brand is missing from AI answers?
  • How do I know whether an AEO fix worked?

Direct answer

Don't work the list in the order the report shows it. Sort each gap by type first, because an absence, a wrong description, a missed recommendation, someone else's page cited in your place, and a page crawlers can't reach or read each need a different fix. Then read the sources each answer cited: they show whether the work is on your page, on a third-party source, or in your site's setup.

Most reports collapse into a handful of causes. Rank them by buyer value times feasibility, write each up as a work item with an owner, an approver, and the facts it rests on, then ship it and judge it over a window of runs, never one rerun.

Step 1: Sort the gaps by type

A gap count mixes unrelated problems. For routing work, sort each gap by where its fix lives; five groups cover most reports, and two are easy to misfile: cited but not recommended is a recommendation problem, and a wrong description only shows up on questions you grade against your real facts. The answer gap brief covers what a content brief should contain.

Gap typeWhere to lookTypical fixHow to verify
Absent from a buyer promptPages and domains cited on that prompt, and whether they mention youEarn coverage in those sources; add your own page if none of yours answers the questionMention rate on that prompt, per engine, across runs
Present but described wronglyThe cited source carrying the wrong fact: often your outdated page, a review profile, or a directoryCorrect the fact there and in every other copyWhether answers still state the old fact
Cited but not recommendedWhich brands get the pick, and which sources name themThird-party coverage and fair comparisons that back your claimRecommendation rate, not citation count
Someone else's page cited where yours could beThe cited URL next to your closest pageAnswer the question as directly and specifically on your pageWhether your URL appears among cited sources
Technical blockServer and CDN logs, bot rules, robots.txt, HTML without JavaScriptAllow the engine's search crawler; put key facts in server HTMLCrawler requests return 200, then the page gets cited; a robots.txt change can take about 24 hours to reach ChatGPT search

Step 2: Read what each answer is built from

For each gap, list the pages and domains cited on that prompt, engine by engine. That list is your work order. If your own page is cited and says the wrong thing, fix your site. If the cited pages are roundups, review profiles, or threads that don't mention you, the work is off-site; a citation gap audit ranks those sources.

If one engine never cites pages the others do, test for a crawler block before rewriting anything; Can AI agents use your website? shows how.

Then group. Gaps that trace to the same source are one problem: a stale pricing page can cause wrong answers on a dozen prompts, and that's one work item.

Step 3: Rank by buyer value times feasibility

How many gaps a cause produces says little about its worth. Score each cause from 1 to 3 on two scales and multiply. Work the 9s and 6s first.

  • Buyer value. 3 for questions asked while choosing: comparisons, pricing, best-for-this-use-case. 2 for category and how-to prompts. 1 for definitions and prompts you already win.
  • Feasibility (how soon you can ship it). 3 for changes you control and can ship this month: your pages, crawler settings, your review-site profiles. 2 for third-party sources that accept corrections and for community coverage. 1 for editorial coverage that takes months. Skip a competitor's own page; you can't change it.

Step 4: Write each fix as a reviewable work item

Each work item needs the 4 parts below, plus the evidence behind it: prompts, answers, cited sources, and date.

  • What to change. The exact edit: a sentence, table, redirect, or setting.
  • Where. The URL or source, plus the prompts and engines it should affect.
  • Who approves it. A named owner and approver: product or legal for claims, the web team for technical changes.
  • The facts it rests on. The approved source for every claim, so the approver checks facts, not wording.

In SolCrys

A finding goes into the Action Hub as a task with its evidence attached, and a Content Audit lists recommended actions for a page; Inside the Action Hub covers the review.

Step 5: Ship, rerun, and judge over a window

Once the change is live, keep running the same prompts on the same engines and compare a window of runs before it with a window after, allowing time for engines to recrawl. Use a 7-, 14-, or 30-day window on each side, the same windows how many runs you need recommends. SolCrys reruns tracked prompts on a schedule, daily by default on paid plans.

One rerun proves nothing: answers are sampled, so a single rerun can show a gain that isn't there or hide one that is. See why your score moves. Also check whether prompts you didn't touch moved on the same engine; if they did, that points to the engine, unless one source or site-wide change you made touches them too.

Log each item: ship date, URL, target prompts and engines, both windows, and the untouched prompts compared. Over a quarter, the log shows which fixes move answers in your category.

Worked example: 41 gaps, 4 causes

Illustrative scenario only. Larchmont, a mid-market observability vendor, gets a report with 41 gaps across 30 buyer prompts on ChatGPT, Google AI Overviews, Gemini, Perplexity, and Claude. Traced to their cited sources, they come down to 4 causes.

  • Wrong pricing, 11 gaps (3 × 3 = 9). Answers say Larchmont charges per host. It moved to usage-based pricing, but an old pricing FAQ is still live and cited, and a review profile copied it. Item: retire and redirect the FAQ, update the profile. Approver: the pricing owner.
  • Blocked on ChatGPT, 9 gaps (2 × 3 = 6). ChatGPT never cites the docs that other engines cite; a bot rule challenges ChatGPT's search crawler, OAI-SearchBot. Item: allow the crawler. Approver: the web team.
  • Absent on Kubernetes logging prompts, 14 gaps (3 × 2 = 6). The answers cite 2 roundups whose vendor lists predate Larchmont's Kubernetes support; both take vendor updates. Item: send both editors the current feature list. Approver: product marketing.
  • Cited, not recommended, 7 gaps (3 × 1 = 3). Larchmont's comparison page is cited on Datadog comparison prompts while Datadog gets the pick. Logged and watched, not assigned.

What moved

The bot rule changes in a day and the pricing fix ships within a week. After 2 weeks of daily runs, pricing answers stop saying per-host on 3 of the 4 engines that said it; the fourth still cites the review profile, whose update awaits approval. ChatGPT starts citing the docs while Larchmont's untouched ChatGPT prompts hold steady.

Sources

FAQ

What if an answer shows no cited sources?

Then there's no page to edit for that run. Rerun the prompt a few times and work from any sources that appear. If none ever do, score the gap low on feasibility.

What do I do with a gap I can't fix?

Log it with the date and the source behind it, and keep it out of the work queue. Recheck it each quarter, because the sources behind an answer change.

How many work items should a small team run at once?

3 to 5 that reach shipped and remeasured teach more than 20 half-done ones. Don't run two items on the same prompts at once, or you can't tell which one moved the answer.

Free ChatGPT visibility check

See where AI answers skip your brand — then fix it, free

Start a free workspace with your domain: 10 buyer-intent prompts through ChatGPT show where you are mentioned, cited, or skipped, and who gets recommended instead. A free content audit in the same workspace hands you the first fix to ship.

Start Free

Free · No credit card · About 5 minutes

Related guides

Citation & Source Influence

Citation Gap Audit

A 5-step framework to identify which sources AI engines cite for your competitors but not for your brand, and the recovery actions for each gap type.

Citation & Source Influence

Cited but Not Recommended: Why AI Citations Are the Wrong KPI

Getting cited as a source and getting recommended as the pick are two different AI-search outcomes, and they are decoupling. Why most teams track the wrong one, what a 100-query study and our own 5-engine data show, and the KPI to measure instead.

Measurement

Why Your AI Visibility Score Moves

AI visibility scores move week to week even when nothing you control has changed. Here are the 6 sources of normal variance, the patterns that signal real movement, and what to watch.