# The AI Search Visibility Audit I'd Run Before Spending Money on GEO

> Run this one-afternoon AI search visibility audit before you pay for GEO services, and set a baseline you can hold any vendor accountable to.

Canonical URL: https://adityamallah.com/blog/ai-search-visibility-audit-before-geo
Author: Aditya Mallah
Published: 2026-10-11T00:00:00.000Z
Updated: 2026-10-11T00:00:00.000Z
Topics: AI search, GEO, AI visibility, SEO, measurement, AI audits

Someone just pitched you a six-month GEO contract. Before you sign anything, spend one afternoon running your own AI search visibility audit: a fixed set of buyer questions, asked across ChatGPT, Gemini, and Perplexity, scored for mentions and citations.

**GEO** is generative engine optimization, paying someone to make your brand show up more often in AI-generated answers. It's a real discipline, but it's also the most expensive thing you can buy blind, because without a baseline you can't tell whether the work did anything.

That baseline costs you an afternoon and about $0. OpenAI said ChatGPT passed a billion weekly users in August 2026, so the traffic question is settled; the question is whether you're visible in those answers and how you'd know if a vendor changed that.

## Why the baseline comes first

You can't measure a vendor's impact without a "before" snapshot. And because AI answers are noisy, one casual check isn't enough. You need a structured snapshot you can repeat.

In January 2026, Rand Fishkin's team at SparkToro, working with Patrick O'Donnell of Gumshoe, ran 12 purchase-intent prompts 2,961 times through ChatGPT, Claude, and Google AI with 600 volunteers. The identical brand list came back less than 1 time in 100 runs, and the identical ordering less than 1 time in 1,000. That's how random a single AI answer is.

But the same study found a stable layer underneath: the top three brands still appeared in 55–77% of responses for a given prompt. In other words, rank position is noise, but inclusion frequency is a genuine signal. That's the whole argument for a baseline built on repeated runs rather than one screenshot.

Now look at what vendors charge. First Page Sage published GEO and AEO pricing statistics in October 2026, built on 1,146 quotes from 214 agencies collected between July and early October. The median retainer is $6,850 a month, up 66% since Q1 2024, and GEO now runs roughly 45% above traditional SEO pricing.

Starter engagements for small businesses sit at $1,800–$4,200 a month on three-month minimums. Growth-tier work for mid-market companies is $4,500–$11,000. Even a standalone one-time audit runs a median of $4,300.

At that money, "trust our dashboard" isn't accountability. Your own baseline is. If a vendor moves your mention rate from 8% to 15% on the same prompt set, that's evidence. If there's no baseline, every claim is unfalsifiable, in their favor.

A baseline also changes the buying conversation before it starts. When you can say "we're mentioned in 11 of 120 category answers on Perplexity, and cited twice," a salesperson can't open with a scary-looking score from their own index. You've defined the unit of measurement, which is exactly the power you give up when you let a vendor define it.

And practically: the audit tells you what to buy. Maybe it reveals your problem is one broken review profile, not a strategy gap. Plenty of audits end with "you don't need us yet", those are the honest vendors, and you'll want them on your side when you do.

## The seven-step AI search visibility audit

Set aside an afternoon for the audit.

### 1. Define your buyer prompt set

Write down 25–50 questions a real buyer would ask before choosing something like yours. Keep the wording verbatim and never edit the set mid-program, or your month-over-month numbers stop meaning anything.

Cover four intent types:

- **Category:** "What are the best CRM tools for a 20-person B2B SaaS team?"
- **Comparison:** "HubSpot vs Pipedrive for a small sales team, which is better?"
- **Best-of:** "Top 10 AI meeting recorders for sales teams"
- **Problem:** "How do I stop leads from going cold after demos?", then note which tools get named

Mix in the phrasings your customers actually use, including messy ones. A prompt that sounds like a real person beats a polished keyword.

### 2. Run the set across the major assistants

Run every prompt in ChatGPT, Gemini, Perplexity, and Copilot, free tiers are fine for this. Use a fresh chat for each run, paste the exact wording, and repeat each prompt three times across different days if you can.

Single runs are meaningless; that's the SparkToro finding. For Google specifically, check AI Overviews separately, since it behaves differently from the chatbots. Ahrefs' free AI Visibility Checker gives you a quick preview without any setup.

Spread the runs over a few days rather than firing all 150 in one sitting. Model behavior shifts with server-side conditions, and spacing the runs out captures a fairer slice of that randomness instead of one unlucky hour.

### 3. Score mentions and citations

**AI share of voice** is your brand's mentions divided by all brand mentions across your prompt set, expressed as a percentage. Calculate it per engine, because they'll differ wildly.

Count citations separately from mentions. Being named in an answer is one thing; being the linked source is another, and it usually means the engine trusts a specific page of yours.

### 4. Log which third-party sources the engines cite

For every answer, record the domains being cited for your category. Rampiq's B2B research, published in June 2026, found roughly 85% of citations for broad category queries come from third parties (G2, Capterra, and TrustRadius were the big three) not from brand websites.

That list is your real to-do list. If every answer in your category leans on one comparison site and one Reddit thread, that's where your next month of effort goes, GEO agency or no GEO agency.

### 5. Audit your entity consistency

Check whether your name, category, and key facts match everywhere the engines looked: your site, LinkedIn, the review platforms, any directories. Models resolve contradictions by guessing, and guesses go to whoever is described most consistently. One site says you're "enterprise workflow software," your G2 profile says "project management," and your LinkedIn says "AI operations platform", the model picks one or invents a blend, and none of them are the story you wrote.

Fix the cheap stuff yourself, an outdated category on a directory, a conflicting product description. This is the least glamorous step and often the highest-leverage one.

### 6. Check crawlability and extractability

Confirm AI crawlers like GPTBot, ClaudeBot, PerplexityBot, and Google-Extended aren't blocked in your robots.txt or firewall, and that your key pages render server-side rather than depending on client-side JavaScript. Then look at the pages themselves: clear headings, direct answers, tables and lists.

The same Rampiq research found structured tables get cited about 2.5 times more than unstructured equivalents, and content refreshed within the last two months earned roughly 28% more citations. Extractable beats beautiful.

### 7. Write the baseline down and repeat monthly

Same prompts, same engines, same scoring, one spreadsheet tab per month. Three months of consistent data tells you more than any vendor's one-off report, and it tells you for free.

## The baseline log: what to record and why

Your audit is only as good as what you capture. These are the columns I'd keep:

| What to record | Why it matters |
|---|---|
| Exact prompt wording, 25–50 prompts | Changing wording invalidates every trend line |
| Engines and models used (and date) | Answers differ per engine and shift after model updates |
| Runs per prompt (e.g., 3) | Single runs are noise, identical lists repeat under 1% of the time |
| Mention: yes/no, and position in any list | Inclusion frequency is the stable signal to trend |
| AI share of voice per engine | One number your whole team can understand |
| Citations: which of your URLs were cited | Shows which pages the engines actually trust |
| Third-party sources cited for your category | Your off-site target list for the next quarter |
| Accuracy of how you were described | Wrong facts need fixing before any promotion |

## What the tools cost

You can run the entire audit manually with a spreadsheet and free chats. Paid LLM monitoring tools just make the monthly repeat less tedious. Here's the starting price of each one I'd actually shortlist.

| Tool | What it does | Starting price |
|---|---|---|
| Manual (spreadsheet + free chats) | Everything in this audit | $0 |
| Ahrefs AI Visibility Checker | Quick free preview of AI answers citing you | Free |
| Otterly.ai | Prompt tracking across ChatGPT, AI Overviews, Perplexity, Copilot | $29/mo for 15 prompts; Gemini and Claude are paid add-ons; free trial |
| Peec AI | Visibility and competitor benchmarking, unlimited seats | $95/mo for 50 prompts across 3 active models |
| Semrush | Full SEO suite with AI visibility built in | Free plan; paid plans from $139/mo, 50 tracked prompts daily at the $199 tier |
| Ahrefs Brand Radar | Custom prompt tracking plus an index of 519M real prompts | $50/mo for custom prompts; $199/mo for the full index; free daily quota on paid Ahrefs plans |
| Profound | Enterprise AEO platform, up to 9 answer engines | Free 7-day trial (50 prompts/day); custom enterprise pricing |

If you're a small team, Otterly's $29 tier or Peec's $95 tier covers the monthly repeat comfortably. Profound's trial is worth a week just to see how an enterprise team structures prompt sets.

## Using your audit to judge a vendor pitch

With a baseline in hand, a pitch turns from a demo into a conversation you can win. There's real evidence GEO work can move numbers, the Princeton-led GEO paper, published at KDD in 2024, measured visibility gains of up to 40% from specific content changes. The key phrase is "specific changes," and your job is to make the vendor name theirs.

Ask these, in this order:

1. **"What's our current visibility, and how did you measure it?"** If they haven't run your prompts, they've already skipped the step that makes their work provable.
2. **"How many runs per prompt, how often, on which engines?"** Good answer: repeated runs on named models, averaged, with raw outputs you can export. Vague answer: "our proprietary index."
3. **"How will you separate your effect from model updates?"** Models get refreshed constantly and answers shift on their own. A serious vendor acknowledges this and designs measurement around it.
4. **"What will you actually change?"** Listen for concrete items: entity fixes, content restructures, specific third-party platforms. Be wary if the plan leans on posting to review sites or mass-producing listicles.
5. **"Show me a client report, raw data included."** You want exports of actual runs, not just a trend chart that starts the month they took over.

Walk away from anyone who guarantees placement in ChatGPT or Perplexity answers. Nobody controls model outputs, and a guarantee is either ignorance or a metric they define after the fact. Also be careful with the contract minimums, three to six months is standard per the First Page Sage data, which is exactly why the baseline has to exist before month one.

## FAQ

**How do I audit my AI search visibility?**
Build 25–50 buyer prompts, run each three times across ChatGPT, Gemini, Perplexity, and Copilot, and log mentions, citations, and cited sources per engine. Compute your AI share of voice, note which third-party sites get cited, and repeat monthly with identical wording.

**What should a GEO audit include?**
At minimum: baseline visibility measurement, a citation source analysis, entity consistency checks across your site and third-party platforms, a crawlability review for AI bots, and a competitor comparison on the same prompts. Anything less is a content review wearing a new label.

**How do I know if GEO services are worth paying for?**
Run the audit first. If your baseline shows near-zero visibility and the cited sources are things you can realistically influence, paid help can speed things up. If you're already cited and consistent, you mostly need content maintenance, which you can do in-house.

**What baseline should I set first?**
One fixed prompt set, four engines, three runs per prompt, scored for mentions and citations, recorded in a spreadsheet with dates. That's the minimum dataset that survives contact with AI answer randomness.

**How much do GEO services cost?**
Per First Page Sage's October 2026 report, median retainers are $6,850/month overall, with starter tiers at $1,800–$4,200, growth tier at $4,500–$11,000, and one-time audits around $4,300. Monthly tools for self-serve monitoring start around $29.

**How long does it take to see AI answers change?**
Faster than SEO in some ways, slower in others. Retrieval-based answers (AI Overviews, Perplexity) can shift within weeks once sources change, while chatbots drawing on trained knowledge move in steps whenever models update, which is why monthly re-measurement matters more than daily checking.

## Sources

- SparkToro (NEW Research: AIs Are Highly Inconsistent When Recommending Brands or Products (Jan 2026)) https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/
- First Page Sage (GEO & AEO Pricing Statistics: 2026 Report (Oct 2026)) https://firstpagesage.com/seo-blog/geo-aeo-pricing-statistics/
- Demand Gen Report (Rampiq Reveals What Content AI Search Engines Cite Most in B2B (Jun 2026)) https://www.demandgenreport.com/industry-news/news-brief/rampiq-reveals-what-content-ai-search-engines-cite-most-in-b2b/53486
- ACM KDD '24 (GEO: Generative Engine Optimization, Aggarwal et al. (2024)) https://dl.acm.org/doi/10.1145/3637528.3671900
- OpenAI (Expanding Access to AI with ChatGPT Ads (Aug 2026)) https://openai.com/index/expanding-access-to-ai-with-chatgpt-ads
- Peec AI (Pricing) https://www.peec.ai/pricing
- OtterlyAI (Pricing) https://otterly.ai/pricing/
- Profound (Pricing) https://www.tryprofound.com/pricing
- Ahrefs (Brand Radar) https://ahrefs.com/brand-radar
- Semrush (SEO + AI Search Pricing) https://www.semrush.com/pricing/

## Related reading

- [How to Check Whether ChatGPT, Gemini, and Perplexity Recommend Your Business](/blog/check-if-chatgpt-gemini-perplexity-recommend-your-business)
- [Why Your Competitors Show Up in AI Search Results and You Don't](/blog/why-competitors-show-up-in-ai-search-results)
- [What Should an SEO and AI Search Agency Deliver in Its First 90 Days?](/blog/seo-ai-search-agency-first-90-days-deliverables)

Source: https://adityamallah.com/blog/ai-search-visibility-audit-before-geo
