AI Visibility Action Plan: Turn a Report Into a Content Plan

An AI visibility action plan turns tracking results into a ranked list of work: each buyer question where AI engines name rivals instead of you becomes a page or outreach task, scored by how many answers it could win back. On our own report, 35 unbranded questions had gaps, adding up to 259 missed answers across 7 topics, and the top three actions covered most of them.

A report full of percentages doesn't tell anyone what to do on Monday. By the end of this post you'll know how to rank actions so the order can be defended, how to turn the top ones into content briefs and a calendar, which plan items to question before acting, and how to tell afterwards whether anything moved. The script that builds the briefs is included.

Data used: 400 answers from BugViso's AI Visibility Tracker for bugviso.com (50 buyer prompts × ChatGPT, Gemini, Perplexity and Claude × 2 runs on 2026-10-20, web search on), against five anonymised competitors. The action plan was produced offline by the tracker's own planning and ranking code on those answers, with no model calls. The content briefs come from the tracker's CSV export.

What Goes Into an AI Visibility Action Plan

Three parts of a tracking report drive the plan: the questions you lose, the sources engines cite instead of you, and the wrong things engines say about you. Everything else (index, share of voice, position) tells you how big the problem is, not what to do.

Our own diagnosis showed which input mattered most. We were named in only 20 of 234 answers that cited third-party sources (8.5 %), while 97 % of checkable claims about us were accurate. So the problem was presence, not perception: engines described us correctly when they named us, and they rarely named us.

How to Rank Actions So the Order Holds Up

Score each action by the answers it could win back, weighted by how close those buyers are to purchase and how many engines show the gap. The tracker's ranking multiplies seven factors:

score = lost answers × buyer intent × engines affected × confidence × ease × strategy × learned

The plan keeps the top five confident actions as the main plan, puts the rest under "Later", and calls the top three "This week". Every item carries its reasons, such as "250 answers where you're missing, mostly shortlist questions, 4 of 4 engines, seen in 2 runs". A client or manager can see why item 1 isn't item 4.

Our Real Plan, Ranked

The ranking put our own pages first, the software research ecosystem second and comparison questions third. This is the plan the tracker produced for bugviso.com:

The order surprised us in one place. Fixing inaccuracies, the item most teams would start with, ranked eighth. With 97 % accuracy, it touched only 10 answers. Our AI brand accuracy audit covers those 10 claims, and they were worth fixing, but a plan ranked by answers won back puts them where they belong.

Question These Plan Items Before You Act

A ranked plan is a starting point, and two of our nine items needed a human check before anyone acted on them.

• The review-sites item named the wrong kind of site. It listed two domains as review sites. One is a genuine software directory. The other is a rival audit tool with a free grader, typed as "review" by the citation classifier, so getting "listed" there would mean listing on a competitor. Open every cited page before outreach.
• Generic steps need real inputs. The own-site item suggests backing claims with "named customers and sourced statistics". That only helps if the customers and numbers are real. We don't publish customer names, so for us that step means sourced data, not logos.
• Reference sources have rules. The reference-footprint item suggests a Wikidata item and, only with independent coverage, a Wikipedia article. Wikidata has its own notability policy; check it before creating an item.

None of these are reasons to ignore the plan. They're the reason a person reviews it before it becomes a brief.

Turn the Top Actions Into Content Briefs

Group the lost questions by topic, pick the page type each question shape needs, and list who's named instead, so the writer knows what to study. Running the script below on our tracker export produced seven briefs:

Briefs are ordered by score, not by missed answers alone. Core Web Vitals ranks above technical SEO with fewer missed answers, because every one of its lost questions was missing on all four engines, while three of the technical SEO questions were missing on only two or three. Each brief carries the exact questions it must answer, for example: "my robots.txt allows AI bots but they still get a 403, which audit tool can diagnose this" (purchase intent, missing on 4 of 4 engines).

Most of our briefs called for "best X for Y" pages because most unbranded buyer questions are shortlist questions. That's a finding about the prompt set as much as the brand. If yours are mostly "how do I" questions, the briefs will be guides instead.

Write each page so engines can quote it. Answer the question in the first two sentences, state numbers and limits in text, and name the audience. Google's guidance on helpful, reliable content is the right standard, and SoftwareApplication markup suits product pages when it matches the visible text.

Separate Content Work From Outreach

Not every plan item is a page you write. In our plan, actions 1, 3 and 5 are content on our own site. Actions 2, 4, 6 and 9 happen on other people's pages. Action 8 is corrections, partly on our site and partly on third-party listings.

The third-party work is where most teams stall, because it can't be scheduled like content. Our AI citation gap analysis shows how to read the cited pages and decide which are worth the outreach.

Measure Whether the Plan Worked

Log each action with its date, then compare the runs before and after it, and only call it a result when the move is bigger than the normal day-to-day swing. The tracker's before/after logic follows four rules worth copying even if you measure by hand:

1. Skip the action day. Engines need time to re-crawl, and you rarely know when a change went live.
2. Average a week on each side. The comparison uses up to 7 runs before and 7 after.
3. Wait for at least 3 runs after. Until then the result is "still measuring".
4. Respect the noise. A move smaller than the swing seen before the action is "within normal variation", not a win. Actions logged within 7 days of each other are shown together, because their effects can't be separated.

These rules matter more than they look. In our data, pooled mention rates moved by one mention between two runs on the same day. A dashboard that celebrates every point will celebrate noise.

Build the Briefs From Your Export

This standard-library script reads a prompt-table export (one row per prompt and engine, with the brand-mentioned count and who appeared instead). It scores every lost unbranded question the same way as the ranking above, groups questions into topic briefs with a page type, the rivals to study and the questions to answer, and lays the briefs out on a calendar.

#!/usr/bin/env python3
"""content_plan.py: turn an AI visibility prompt-table export into a ranked content plan.

Usage: python3 content_plan.py prompt_table.csv [--weeks 4] [--per-week 2]

prompt_table.csv: one row per prompt and engine, with columns prompt, kind, intent, topic,
engine, successful_samples, brand_mentioned, appears_instead.
"""
import csv, re, sys
from collections import defaultdict

INTENT = {"purchase": 1.0, "consider": 0.85, "learn": 0.55}
PAGE = [  # (pattern on the prompt, page type to write)
    (r"\bvs\.?\b|\bversus\b|\bor\b.*\bfor\b|compare|alternative", "comparison page (you vs each rival, honest trade-offs)"),
    (r"\bcost|price|pricing|cheap|plan|quota|per month|free\b", "pricing explainer with every plan, limit and price in text"),
    (r"\bbest\b|\btop\b|recommend|\bwhich\b.*\b(tool|software|platform)\b|tool for", "'best X for Y' page naming who you are and aren't for"),
    (r"^how |how do|how to|what is|why ", "guide answering the question in the first two sentences"),
]


def page_for(prompts):
    votes = defaultdict(int)
    for p in prompts:
        for rx, page in PAGE:
            if re.search(rx, p, re.I):
                votes[page] += 1
                break
        else:
            votes["feature page stating the capability by name"] += 1
    return max(votes, key=votes.get)


def main(path, weeks=4, per_week=2):
    rows = list(csv.DictReader(open(path, newline="")))
    by_prompt = defaultdict(list)
    for r in rows:
        if r.get("kind") != "branded" and r.get("engine"):
            by_prompt[r["prompt"]].append(r)
    gaps = []
    for prompt, rs in by_prompt.items():
        samples = sum(int(r["successful_samples"] or 0) for r in rs)
        hits = sum(int(r["brand_mentioned"] or 0) for r in rs)
        missing_engines = [r["engine"] for r in rs if int(r["brand_mentioned"] or 0) == 0]
        if not missing_engines:
            continue
        score = (samples - hits) * INTENT.get(rs[0]["intent"], 0.7) * len(missing_engines) / len(rs)
        gaps.append({"prompt": prompt, "topic": rs[0]["topic"], "intent": rs[0]["intent"], "score": score,
                     "missed": samples - hits, "samples": samples, "engines": missing_engines,
                     "instead": rs[0].get("appears_instead") or ""})
    briefs = defaultdict(list)
    for g in gaps:
        briefs[g["topic"]].append(g)
    ranked = sorted(briefs.items(), key=lambda kv: -sum(g["score"] for g in kv[1]))
    total = sum(g["missed"] for g in gaps)
    print(f"{len(gaps)} unbranded prompts with gaps, {total} missed answers, {len(ranked)} topics\n")
    for i, (topic, gs) in enumerate(ranked, 1):
        gs.sort(key=lambda g: -g["score"])
        rivals = sorted({re.sub(r"\s*\(\d+\)", "", x).strip() for g in gs for x in g["instead"].split(";") if x.strip()})
        print(f"Brief {i}: {topic}  (score {sum(g['score'] for g in gs):.0f}, {sum(g['missed'] for g in gs)} missed answers)")
        print(f"  page: {page_for([g['prompt'] for g in gs])}")
        print(f"  study who's named instead: {', '.join(rivals[:4]) or '-'}")
        for g in gs[:3]:
            print(f"  must answer: \"{g['prompt']}\"  [{g['intent']}, missing on {len(g['engines'])}/4]")
        print()
    print("calendar:")
    slots = [t for t, _ in ranked][: weeks * per_week]
    for w in range(weeks):
        print(f"  week {w + 1}: " + " | ".join(slots[w * per_week:(w + 1) * per_week]))


if __name__ == "__main__":
    a = sys.argv[1:]
    if not a:
        sys.exit(__doc__)
    weeks = int(a[a.index("--weeks") + 1]) if "--weeks" in a else 4
    per = int(a[a.index("--per-week") + 1]) if "--per-week" in a else 2
    main(a[0], weeks, per)

The first brief and the calendar from our export, unedited:

$ python3 content_plan.py prompt_table.csv --weeks 3 --per-week 2
35 unbranded prompts with gaps, 259 missed answers, 7 topics

Brief 1: AI search readiness and crawler access  (score 38, 48 missed answers)
  page: 'best X for Y' page naming who you are and aren't for
  study who's named instead: Competitor A, Competitor B, Competitor C, Competitor E
  must answer: "which GEO audit tool should I use to fix why AI assistants can't cite my site"  [purchase, missing on 4/4]
  must answer: "my robots.txt allows AI bots but they still get a 403, which audit tool can diagnose this"  [purchase, missing on 4/4]
  must answer: "best tool to check if my website is ready to be cited by AI search engines"  [consider, missing on 4/4]

calendar:
  week 1: AI search readiness and crawler access | Core Web Vitals and performance testing
  week 2: SEO and technical website checks | White-label agency reports
  week 3: Pricing and free reports | Audit scan quotas and rollover

(Briefs 2–7 follow the same format and are trimmed here.) The page-type rules are deliberately simple keyword patterns. Read the questions under each brief before you commit to a format.

How BugViso Builds the Action Plan

BugViso's AI Visibility Tracker writes the "what to do" plan from each report automatically, ranks it with the formula above, and drafts the assets for each action with AI. The plan comes from your answers on ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews (web search on, 50 buyer prompts, at least three competitors).

Each item shows its reasons, the questions it targets, the rivals named instead and the pages cited instead of you, so the order can be explained. When you mark an action done, impact tracking compares the runs before and after with the honesty rules in the previous section. The plan, source gaps and accuracy check all sit in the report, which exports as PDF or CSV. More on the AI visibility tracker page.

Mistakes That Stall an Action Plan

• Starting with what's easiest to explain. Accuracy fixes felt urgent and ranked eighth for us. Rank by answers won back.
• Writing one page per question. Group by topic. Seven briefs covered our 35 lost questions.
• Skipping the outreach half. Four of our nine actions happen on third-party pages. Assign them to someone.
• Trusting labels blindly. One "review site" in our plan was a rival's free tool. Open every cited page.
• Measuring the day after. Wait for at least three runs, and compare against the normal swing.
• Changing the prompt set while measuring. Before/after only works on the same questions.

Frequently Asked Questions

What is an AI visibility action plan?

An AI visibility action plan is a ranked list of changes meant to get a brand named more often in AI answers, built from tracking data. Each action targets specific buyer questions where engines name competitors instead, and carries the reasons for its rank: missed answers, buyer intent, engines affected and effort.

How do I prioritise AI visibility fixes?

Score each fix by the answers it could win back, weighted by buyer intent (purchase questions highest), the share of engines showing the gap and how confident you are that the gap is real. Prefer low-effort fixes among equals. In our plan that put our own pages first (250 missed answers) and accuracy fixes eighth (10).

What content gets brands mentioned in AI answers?

Pages that answer the exact buyer question in their first sentences, state numbers and limits in text, and say who the product is for. In our data most lost questions were shortlist questions, so the briefs called for "best X for Y" pages and plain pricing explainers. Being listed on the third-party pages engines cite matters as much.

How long does it take to see results from GEO changes?

Wait for at least three tracking runs after the change before judging, and compare a week of runs on each side, skipping the change day. Engines need time to re-crawl. A move smaller than the normal day-to-day swing is noise, however good it looks.

Can I build an AI visibility plan without a tracker?

Yes, for a small prompt set. Ask each engine your buyer questions with web search on, record who's named and which pages are cited, then group lost questions by topic and rank them by intent and engine count. The script above does the grouping from any export with the same columns.

Rank by Answers Won Back, Then Write

A plan earns trust when anyone can see why item one comes first. Rank by missed answers, intent and engines; question every label; split content from outreach; and measure with honest rules. That's the loop BugViso's AI visibility tracking runs after every report.