AI SEO Agency Red Flags: Twelve Warning Signs, and When Each Shows Up

Why this field attracts warning signs

Getting named in AI answers is new, results are hard to see, and almost nobody buying it has a way to check the claims. That combination attracts both honest newcomers and a smaller number of people who are happy to overpromise. You cannot tell the groups apart by how confident they sound.

This page lists twelve warning signs, what each one looks like in practice and what to do about it. None of them proves bad faith on its own. Some come from a good agency having a poor day or an unpracticed salesperson. But two or three together are a clear signal to step back.

If you are still working out what a fair engagement looks like, start with our guide to what a service should deliver. For the questions that bring flags to the surface, see questions to ask an AI SEO agency.

One note on tone. We will not name any agency here. The flags describe behavior, and behavior is what you can check.

When flags appear, and why that matters

Warning signs appear at four moments, as the diagram shows. The pitch shows the claims. The contract shows what the agency is willing to be held to. The first month shows whether the work matches the pitch. The third month shows whether the reporting is honest.

That timing matters because the cost of acting rises each step. Walking away at the pitch costs an hour. Walking away at the contract costs a conversation. Walking away at month three costs money and time and often an argument.

So read the first six flags below as things to catch before you sign. The last six are things to watch for once work has started, and to raise early if you see them.

Flags 1 to 4: In the pitch

1. A guaranteed mention

Nobody controls which businesses an AI assistant names. The same question can produce different answers on different runs. Any promise of a guaranteed appearance is either a misunderstanding or a misrepresentation. A related version defines "mentioned" so loosely, such as appearing once in one answer, that the guarantee is easy to meet and means nothing.

2. A secret method

"We use a proprietary system we cannot explain" is a bad sign in a field where the basics are well understood. Real methods can be described in plain words: measure, fix access, write answer-first pages, work on outside sources, measure again. If the explanation stays vague when you press, there may be nothing underneath.

3. One screenshot as proof

Assistants vary. A screenshot of one answer naming a client may be typical or may be the best of twenty tries. Ask for a repeated measurement over months instead. If the agency has only screenshots, it has never measured properly.

4. Rankings as proof

A chart showing a client ranking first on Google proves skill at ordinary search. It does not show that any assistant names that client. Ask for evidence from answers, not from rankings. See how to choose an AI search optimization agency for how to read proof.

Flags 5 and 6: In the method

5. Hidden text, prompt tricks and "AI poisoning"

Some providers sell ways to talk to the assistant behind the reader's back: white text on a white background, instructions buried in a page, files designed to steer a model. These tricks are dishonest, they tend to be caught and they risk penalties from search engines and from the assistants' own makers. A site that earns trust by being useful does not need them.

A related trap is a special file offered as a switch. Files that describe your site can be harmless and occasionally helpful. None is a switch, and a proposal that leans on one has little else.

6. Bought mentions and bulk placements

Paying for hundreds of low-quality directory entries, comment links or sponsored posts to "build authority" does not make assistants trust you. They lean on sources that independent sites and people already vouch for. A pile of paid placements in thin venues can be ignored at best and a liability at worst. Real outside work is slow and specific, and cannot be bought by the hundred.

Our page on what AI engines cite shows what kinds of source tend to count.

Flags 7 to 9: In the contract and the plan

7. Activity promised, outcome never mentioned

"Twelve pages a month" tells you what the agency will do and nothing about what it will achieve. Ask what outcome you can hold them to, and whether any part of the fee depends on it. An agency that will agree to nothing measurable is selling hours.

8. No baseline before work starts

Without a measurement before any change, every later improvement can be claimed and every decline blamed on the assistants. A baseline protects you and a refusal to take one protects them. It is the single most important clause.

9. A plan that is only about your own website

Assistants lean on outside sources. A plan that never leaves your domain has dropped the half of the job that often decides who gets named. This is not always sinister, because some agencies only know on-site work, but it is a gap you are paying for.

Also watch for long lock-ins. Twelve months up front for work whose results show slowly asks a lot of a client who cannot yet judge quality. Our page on how much AI SEO costs covers how terms affect the real price.

Flags 10 to 12: Once work has started

10. A question list that keeps changing

If the questions measured change every month, no trend can be read. Sometimes this is innocent, as a team adds new ideas. But if originals are dropped without saying so, the history has been quietly rewritten. Ask for the original questions to keep running unchanged beside any new ones.

11. Traffic as the headline metric

Most recommendations from assistants produce no click. A buyer hears a name and searches for it later, or visits directly. A report that leads with referral sessions will show failure during genuine progress and success during a fluke. Traffic is worth showing as a secondary view, not as proof.

12. Reports with only good news

In a field this variable, some questions move up, some move down and many wobble. A report that shows only gains has been curated. Look for a note on what did not improve, quotes from real answers and results split by assistant.

If you see two of these three, ask for the full raw answers from the last run. A good agency sends them the same day. A poor one has to think about it.

Smaller signs that add up

A few things are not red flags alone but are worth noticing together.

Pressure to sign this week. Real work does not expire on Friday. A discount that disappears at midnight is a sales tactic, and a good one will still be there after you have asked your questions.

Reluctance to speak to the person who will do the work. If you only ever meet a salesperson, you do not know who is behind the service.

Criticizing every other approach. When one agency says only its method works, it usually means it has one method.

Testimonials with no names, no companies and no way to check. Anyone can write a quote.

Avoiding the word "measure". A real practitioner raises it before you do.

Pages produced in huge numbers with little visible difference between them. Hundreds of near-identical pages are the old programmatic trick with a new label. An assistant choosing what to cite wants the best answer, and thin copies are rarely it.

Warning signs that show up in the money

Pricing behavior tells you a lot, and it is easy to check.

A single figure with no breakdown is the first sign. You should be able to see what the fee includes: how many questions are measured, how many assistants, how many pages and what review each page gets. If the answer to "what is in the fee" is "everything," ask which things.

A large setup fee for a small amount of setup is the second. The first-week work, such as a crawler check and a baseline, is real but modest. A big upfront charge for it deserves an explanation.

Extras that appear later are the third. Developer time, extra assistants, extra questions and reporting access are common add-ons. Ask for them in writing at the start.

A package that is wildly cheaper than all the others is the fourth. It usually means thin pages in volume, or software output with no human review. Our page on how much AI SEO costs explains why price follows scope.

What the opposite of each flag looks like

It helps to know the good versions, so you recognize them when they appear.

Instead of a guarantee, a promise about the work and the measurement. Instead of a secret method, a plain description you could repeat to a friend. Instead of one screenshot, a series over months. Instead of rankings, quoted answers.

Instead of activity, an outcome you agree on in advance. Instead of no baseline, a baseline in week one. Instead of a plan about your site only, a plan that names outside sources. Instead of a shifting list, a fixed one with additions tracked separately.

Instead of traffic as the headline, answers split by assistant. Instead of good news only, a note on what did not move. And instead of pressure to sign, a willingness to wait while you check.

Agencies that show these behaviors are usually easy to talk to. They are comfortable being measured because they have been measured before.

What to do when you spot one

You have three options, in order of preference.

Ask about it directly and kindly. "I noticed the proposal does not mention a baseline. How do you handle that?" A good agency welcomes the question and either explains or fixes the proposal. This gives you real information.

Negotiate around it. If the agency is otherwise strong, ask for the missing safeguard in writing: a baseline, a shorter term, ownership of the history, a fixed question list. Most good agencies agree.

Walk away. If the flag is a guarantee, a trick or a refusal to show proof, do not argue. Thank them and leave. The cost of one more call is far lower than the cost of a bad engagement.

It also helps to separate a flag about the agency from a flag about the salesperson. A new salesperson may overstate things the delivery team would never say. If you suspect that, ask to speak with the person who would run the work, and put the same question to them. Their answer is the one that counts, because they are the ones who will be held to it.

Whatever you decide, write down what you saw. If you speak to several agencies, notes let you compare fairly, and they remind you later why you chose as you did.

How RunOctopus handles this

RunOctopus makes no guarantee that an assistant will name you, because no one can. Its AI, Ollie, researches your business, then writes and publishes pages directly on your own website, with a team planning, writing and reviewing alongside it. It works with Shopify, WordPress, WooCommerce and Wix.

You install nothing, approve nothing, upload nothing and schedule nothing, and pages go live on their own. The first week brings 8 long-form articles, 6 pages and a custom interactive tool for your niche, followed by 175+ pieces of content a month.

It reports on where ChatGPT, Perplexity, Claude and Gemini name your business. The promise is more visibility on Google and in AI answers within 90 days, measured against the 90 days before you joined, or every dollar back.

Sterman, a real estate business in Los Angeles, went from 44 to 210 Google clicks a month since April (Google Search Console, 28 days to September 29, 2026).

Check for yourself, free: paste your address for the site read. No account, no call.

The short version

Be most careful of guarantees, secret methods, single screenshots and refusals to take a baseline. Those four tell you most of what you need.