Here is a genuinely uncomfortable exercise for any ERP consulting firm leader. Open ChatGPT, ask it to recommend implementation partners in your exact niche, and read what comes back. Then ask what it knows about your firm specifically.
Most ERP partner executives have never done this. The ones who do usually discover one of three things:
- They are absent from answers their competitors appear in.
- They are present but described inaccurately.
- Occasionally, they are pleasantly surprised and have no idea why.
All three discoveries lead to the same conclusion. Your firm is being represented, daily, in a channel you have never once measured.
An AI visibility audit fixes that. It is not complicated and it does not require special tooling, just a structured afternoon per quarter and the discipline to log what you find. This post is the complete framework we run for ERP consulting firms, laid out so you can run it yourself.
Key takeaways:
- An AI visibility audit measures whether assistants name your firm, describe it accurately, and which sources they lean on when they do.
- The audit has seven steps: build a prompt panel, run it across four assistants, score the answers, trace the sources, check your own entity footprint, benchmark competitors, then prioritize fixes.
- The source-tracing step is the one that converts measurement into a work plan, and the one most DIY audits skip.
- Quarterly is the right cadence. Expect accuracy to improve by the second audit and presence by the third.
What is an AI visibility audit?
An AI visibility audit is a structured, repeatable review of how AI assistants describe and recommend your firm. You run a fixed panel of buyer-style prompts across ChatGPT, Perplexity, Gemini, and Copilot, score each answer for presence and accuracy, trace the sources the assistants relied on, and convert the findings into a prioritized fix list.
Think of it as rank tracking for the answer layer. Classic SEO reporting tells you where you stand on a results page. The audit tells you where you stand inside the answer itself, which is where a growing share of ERP partner selection now begins. As we argued in how AI assistants decide which ERP firms to recommend, these answers are built from inputs you can influence. The audit tells you which inputs, in what order.
Why ERP consulting firms need this now?
Back in early 2024, Gartner predicted that traditional search engine volume would drop 25 percent by 2026 as queries shifted to AI chatbots and virtual agents. Now that 2026 has arrived, the exact number is debatable. The direction is not. A meaningful slice of the questions your buyers used to type into Google is now being asked in a chat box, and answered in a paragraph rather than a results page.
For ERP consulting firms specifically, three things make this channel impossible to ignore:
- ERP selection is a shortlist game, and assistants compress shortlists. A selection committee that once assembled a longlist of ten firms now asks an assistant for three to five names. If you are not in the compression, you are not in the deal, and you never even see the RFP you lost.
- The buyer’s hardest questions are the assistant’s easiest. “Compare Acumatica and NetSuite for field services, and tell me who implements each well” is a miserable Google query and a perfect ChatGPT one. Buyers have figured this out faster than partners have.
- The channel is invisible in your CRM. Nobody writes “ChatGPT told us about you” in the referral field. The influence shows up as unexplained direct traffic and suspiciously well-informed first calls, which is precisely why it needs its own measurement.
If generative engine optimization is new territory for you, start with our primer on GEO for ERP consulting firms, then come back. The audit below is the measurement half of that discipline.
The audit at a glance
Seven steps, run in order, logged in one spreadsheet. Budget half a day for the first pass. Repeat passes go faster because the prompt panel and the scoring sheet already exist.
| Step | What you do | What you get | Time |
| 1. Build the prompt panel | Write 15 to 25 buyer-style prompts across four intent layers | A reusable question set | 1 to 2 hours, once |
| 2. Run the panel | Ask every prompt across ChatGPT, Perplexity, Gemini, and Copilot | A verbatim answer log | 2 to 3 hours |
| 3. Score the answers | Grade presence, position, accuracy, sources, and competitive set | A presence rate and an accuracy grade | 1 to 2 hours |
| 4. Trace the sources | Identify the pages and profiles feeding each answer | A source-level work plan | 1 to 2 hours |
| 5. Audit your entity footprint | Check the inputs you control: site, profiles, schema, crawlers, reviews | A consistency checklist | 1 hour |
| 6. Benchmark competitors | Run the same prompts for your top three rivals | Attribution for what the machines reward | 1 hour |
| 7. Prioritize and schedule | Sort findings by effort and impact | A quarterly fix list with owners | 30 minutes |
Step 1: Build the prompt panel
The audit is only as good as its questions, and the questions should be your buyers’, not yours. Build a set of fifteen to twenty-five prompts across four intent layers, drawn from the same journey mapping we covered in the ERP buyer journey.
| Prompt layer | What it simulates | Example for an ERP consulting firm |
| Recommendation | The shortlist forming | “Recommend a NetSuite implementation partner for a mid-market manufacturer in the Southeast.” |
| Comparison | Head-to-head evaluation | “Compare [your firm] and [competitor] for a Dynamics 365 Business Central implementation.” |
| Direct knowledge | Diligence on your firm | “What do you know about [your firm]? Is [your firm] a good NetSuite partner? Who are their typical clients?” |
| Situation | The messy, realistic buyer | “We are a 90-person distributor on QuickBooks with a warehouse in Ohio, evaluating NetSuite. What should we look for in a partner, and who should we consider?” |
A few rules keep the panel honest:
- Vary industry, company size, geography, and platform across versions. A NetSuite Solution Provider, an SAP Business One reseller, and an Acumatica VAR should each write prompts that mirror their own pipeline, not a generic one.
- Weight the panel toward recommendation prompts. These are the money questions. Shortlists live here.
- Do not skip the situation prompts. The long, messy, realistic ones are where assistants shine, and buyers know it. They are also where accurate positioning wins or loses, because the assistant has to match your specialty to the scenario.
- Write the prompts once and freeze them. Keep them verbatim in a spreadsheet and reuse them every quarter. Consistency is what turns observation into trend.
Step 2: Run the panel across the assistants
Run every prompt across the major assistants. They retrieve differently, cite differently, and disagree often, which is exactly why one-model audits mislead.
| Assistant | Why it is in the panel | What to note |
| ChatGPT | The default assistant for most buyers, with the largest usage base | Sources appear when it browses; answers vary between runs |
| Perplexity | Citation-first by design | The clearest window into which pages feed answers |
| Gemini | Retrieval tied to Google Search | Tracks your classic Google rankings most closely |
| Copilot | The default inside Microsoft-centric companies, which describes most ERP buying committees | Grounded in Bing, so your Bing indexing suddenly matters |
Method matters more than it seems:
- Use fresh sessions so earlier conversation does not contaminate the answers.
- Run the important prompts twice. Generated answers vary, and you care about the stable signal, not one roll of the dice.
- Do not run the audit logged in as yourself where the platform personalizes. You want the buyer’s view of your firm, not your own reflection.
- Capture everything verbatim. Full answer text, cited sources, screenshots for the record.
The trade press that tracks this space, Search Engine Land among them, has documented how quickly answer behavior shifts between model updates. Your verbatim log is how you will know whether a change next quarter was your doing or theirs.
Step 3: Score what came back
Turn the transcripts into numbers. Five metrics per prompt, per assistant:
| Metric | The question it answers | How to record it |
| Presence | Did your firm appear at all? | Yes or no |
| Position | First mention, mid-list, or afterthought? | 1 to 3 |
| Accuracy | Is the description right on specialty, platform, geography, and size? | Letter grade |
| Sources | What did the assistant cite or lean on? | List of URLs |
| Competitive set | Who else was named, and how were they framed? | Names plus one-line framing |
Score accuracy honestly, because flattering errors are still errors. An ERP firm described as “strong in manufacturing” when the bench is actually professional services will attract the wrong RFPs and quietly lose the right ones. Misrouted demand costs more than absence, because it also burns sales time.
Two composite views make the spreadsheet useful:
- Presence rate. The share of recommendation prompts where your firm was named, overall and per assistant.
- Accuracy grade. A letter grade for how the assistants describe your firm when asked directly.
Between them you get the headline every quarterly review needs: are we in the answers, and are the answers right?
Step 4: Trace the sources
This is the step that converts measurement into a work plan, and the one DIY audits most often skip.
For every answer that named you, or conspicuously did not, look at what fed it. Perplexity shows citations natively; the others reveal sources when browsing. Patterns emerge fast. Recommendation answers in the ERP world lean repeatedly on a familiar cast: review platforms like G2 and Clutch, vendor partner directories, “top partner” listicles, community threads, and the specific firm pages that rank for the underlying query.
Each pattern is an instruction:
| What the source trail shows | What it means | What to do about it |
| A cited listicle you are absent from | A load-bearing page in your niche | An outreach target, this month |
| A directory profile feeding a stale description | Third-party profiles are outvoting your homepage | An afternoon’s fix, made at the source |
| A competitor’s case study cited for your exact specialty | Citation magnets work in your niche | A template for your response |
| Your service page ranking but never cited | An extractability problem, not a substance problem | Clearer claims, better structure, schema |
| Community threads naming your rivals | Word of mouth the machines can read | Earn mentions where your buyers actually talk |
That third row deserves emphasis. When a rival’s case study is being quoted back to buyers for exactly the work you do better, you are looking at live proof that the citation magnet playbook works in your niche, along with a free blueprint for your counter-move.
Step 5: Audit your own entity footprint
With the outside view logged, turn the lens inward and check the inputs you directly control, in one sitting:
| Check | Where to look | Pass condition |
| Plain-language positioning | Homepage and about page | A stranger, or a machine, can state what you do, for whom, and where in one sentence |
| Description consistency | LinkedIn, Crunchbase, vendor directories, review profiles | The same firm with the same claims everywhere, not three slightly different firms |
| Structured data | Your key pages | Valid markup in the Schema Markup Validator |
| Crawler access | robots.txt, CDN settings, security plugins | GPTBot, PerplexityBot, ClaudeBot, and Google-Extended are not blocked |
| Reviews | The platforms the answers actually cite | Recent, real, and where the machines are looking |
| Rankings under the prompts | A rank tracker like Semrush | You rank for the queries beneath your recommendation prompts |
You are not chasing Wikipedia-grade notability here. You are making sure every rung of the corroboration ladder you can reach is actually climbed. The schema row alone fixes a surprising number of description errors; the full build order is in our practical guide to schema markup for consulting firms. And the review row does double duty, because the same trust signals that persuade a human selection committee are the ones assistants quote back to buyers.
Classic SEO telemetry belongs in this step too. Rank tracking for the queries underneath your recommendation prompts tells you whether the retrieval layer can even find you, which is the same ground our NetSuite Partner SEO work covers, and the reason GEO never fully replaces SEO. Your analytics will also increasingly show referral traffic arriving from the assistants themselves, a small but delightfully concrete measure of the channel working.
Step 6: Benchmark your competitors
Run the direct-knowledge prompts for your top three competitors and score them on the same sheet. The point is not envy. It is attribution.
When a rival is consistently named and accurately described, their source trail shows you which inputs the machines are rewarding in your niche right now, with the guesswork removed. Which directories feed their description. Which case studies get cited. Which listicles carry them into recommendations. Reverse-engineering a competitor’s citations is the closest thing GEO has to reading the answer key.
Step 7: Prioritize and schedule the fixes
You will finish the audit with more findings than time. Sort them on two axes, effort and impact, and the top of the list is remarkably consistent across ERP consulting firms:
| Fix | Effort | Impact | Typical timing |
| Unblock AI crawlers | Ten minutes | High | Today |
| Correct wrong descriptions at their source | Low | High | This week |
| Complete and refresh directory and review profiles | Low | High | This month |
| Add missing schema | Low to medium | Medium to high | This month |
| Citation magnet content | High | High, compounding | This quarter |
| Review generation program | Medium | High, compounding | Ongoing |
| Ranking improvements on the underlying queries | High | High, compounding | Ongoing |
The first four rows are days of work, not quarters. Behind them queue the compounding builds: citation magnet content, review generation, the mention program, ranking improvements on the queries beneath the prompts, the whole GEO and SEO system working together rather than competing for budget.
Then put the next audit on the calendar. Quarterly is the right cadence: frequent enough to catch model shifts and measure your fixes, infrequent enough to let changes propagate. Firms that run this loop stop treating AI answers as weather and start treating them as what they are, a channel inside the broader ERP lead generation system, with inputs, outputs, and a trend line.
The five surprises the first audit usually delivers
After running this for enough ERP consulting firms, the findings cluster into a familiar top five. Worth knowing in advance, so they sting less.
- You are described by your oldest self. The machines confidently recite the positioning from a directory profile nobody has touched since 2021: the wrong specialties, the abandoned service line, the city you moved from. Stale third-party profiles outvote your current homepage more often than seems fair, which is exactly why the source-tracing step exists.
- One listicle is doing enormous work. A single “top partners” article, sometimes years old, sometimes mediocre, turns out to feed a startling share of recommendation answers in a niche. Absence from it is expensive; presence in it is cheap by comparison. Every ERP niche has one or two of these load-bearing pages, and the audit finds yours.
- Your best proof is invisible. The case study that would perfectly answer the buyer’s question exists, but as a PDF, or behind a form, or on a page with no extractable summary, and the machines cite a competitor’s thinner but readable page instead. Format, not substance, is the gap.
- The assistants disagree about you. Named warmly by one model, unknown to another. This is normal, it maps to their different retrieval habits, and it is genuinely useful: the model that knows you shows which sources work, and the model that does not shows which are missing.
- Somebody blocked the crawlers. A security plugin, a CDN default, a well-meaning IT decision from 2023. The fix takes ten minutes and the discovery justifies the whole audit.
None of these are disasters. All of them are invisible until you look, and every one converts directly into a line on the fix list, which is the entire point of looking.
Turning the audit into a report someone reads
A spreadsheet convinces the person who built it. A one-pager convinces the person who funds it. After each quarterly pass, compress the findings into four blocks:
- The headline numbers. Presence rate and accuracy grade, this quarter versus last, per assistant.
- The competitive line. Who the machines name most in your niche, and where you sit.
- Three fixes shipped, and what moved because of them. This is how the program earns its next quarter.
- Three fixes queued, with owners and effort attached.
One page, ten minutes to read, and it reframes the whole exercise from curiosity to channel management. Firms that report this way stop having the “is AI visibility real” debate internally within about two cycles, because the trend line answers it for them.
What good looks like, quarter by quarter
| Quarter | Realistic expectation | The signal you are watching for |
| First | A baseline, and a few unflattering discoveries | You finally know what the machines say about you |
| Second | Accuracy improves | Corrected descriptions propagate from their sources into answers |
| Third | Presence improves | Profiles, reviews, and citation magnets start feeding recommendations |
| Fourth and beyond | A defensible trend line | AI answers managed as a channel, not checked as a curiosity |
One caution on timing: propagation is not instant. A description corrected at its source can take weeks to show up in answers, because the assistants have to re-crawl, re-index, and in some cases re-train their retrieval layer on the updated page. Resist the urge to declare failure at week three. The quarterly rhythm exists precisely so that fixes have room to land before you measure them.
Frequently asked questions about AI visibility audits
How often should an ERP consulting firm run an AI visibility audit? Quarterly. That cadence is frequent enough to catch model updates and measure the effect of your fixes, and infrequent enough that changes have time to propagate through the assistants’ sources. Monthly audits mostly measure noise; annual audits let a wrong description misroute buyers for a year.
Do we need special software to run one? No. The first pass needs the assistants themselves, a spreadsheet, and discipline. A rank tracker helps with the retrieval-layer checks, and enterprise AI visibility platforms exist for firms running large prompt panels, but neither is a prerequisite. The framework in this post is deliberately tool-agnostic.
Which AI assistants should the audit cover? At minimum ChatGPT, Perplexity, Gemini, and Copilot. They retrieve and cite differently, so a one-model audit gives a false picture. Copilot deserves particular attention in the ERP world, because Microsoft-centric buying committees, which is most of them, meet it by default inside the tools they already use.
What is a good presence rate? There is no universal benchmark, and anyone quoting one is guessing. The useful comparisons are your own baseline and your top three competitors on the same prompt panel. A boutique firm being named in a meaningful share of its niche recommendation prompts, and moving that share quarter over quarter, is winning regardless of the absolute number.
How is this different from an SEO audit? They overlap but answer different questions. An SEO audit measures whether search engines can find, crawl, and rank your pages. An AI visibility audit measures whether answer engines name your firm, describe it accurately, and which sources they trust when they do. The retrieval layer connects them, which is why the audit includes classic rank tracking, but neither substitutes for the other.
Can a small ERP partner realistically compete here? Yes, and often faster than in classic SEO. Answer engines reward specificity, and a twelve-person firm with a sharp niche, consistent profiles, and extractable proof frequently outperforms a generalist ten times its size in recommendation prompts for that niche. The audit is how you find out which specific inputs your niche rewards.
The version where we do it
Everything above is genuinely DIY-able, and we wrote it so you can. The done-for-you version, the audit we run inside our NetSuite Partner GEO engagements, adds scale and interpretation: larger prompt panels, full competitor sets, source-trail analysis across every finding, and a prioritized fix list with the work estimated. Either way, the first step is identical and free: ask us for the audit, or open the assistants tonight and ask them about your firm.
Just do one of them soon. The machines are already answering questions about ERP consulting firms like yours. The only choice you have is whether you know what they say.
