Video: Transform AI Speak into Evaluator-Ready Writing | Duration: 3608s | Summary: Transform AI Speak into Evaluator-Ready Writing | Chapters: Webinar Introduction (29.39s), AI Speak Pitfalls (103.905s), AI in Proposals (185.28499s), AI in Proposal Writing (272.07s), AI in Proposals (469.36002s), AI Speak Experiment (566.065s), AI Speak Problems (706.605s), AI Text Limitations (876.86s), AI Speak Pitfalls (986.075s), Identifying AI Speak (1129.975s), Enhancing Proposal Quality (1404.415s), Enhancing AI Creativity (1713.0199s), Mitigating AI Speak (1962.83s)
Transcript for "Transform AI Speak into Evaluator-Ready Writing":
Welcome everyone to today's Deltek Thought Leadership Webinar. I'm Noor Bashoor, Director of AI Proposal here at Deltek. Before we get started today, we've got just a few quick reminders for you. For the best webinar experience, we recommend you use Google Chrome or Firefox. You can resize the widgets on your screen and if you've got any questions, please type them in the Q and A box on your screen. We've got a lot of great content for you today, so if we can't get to every question during the webinar, they'll be answered offline afterwards. All of the resources from today's webinar, including today's slide deck, can be downloaded in the Resources widget, which should be in the upper right corner of your screen. In case you need to leave early or re listen to a section of today's webinar, you will receive a link to the on demand recording of today's webinar via email within twenty four hours after the webinar ends. Today, Deltek is pleased to partner with Lowfell Consulting Group and our continued efforts to bring you actionable intelligence and strategies you can use right away. Today, our discussion topic is transform and speak into evaluator ready writing. Our expert presenters are Beth Wingate and Brenda Chris. Beth and Brenda, the floor is yours. Thank you so much. Proposal evaluators can spot AI speak from a mile away. During this session, we'll discuss the pitfalls of AI speak, show you how to identify it in your proposals, provide practical methods to transform it into higher scoring, customer focused writing. AI speak is one of four major risks associated with using AI responsibly the others are security hallucination and bias and we discuss these risks in in our low fell consulting website we've also created a free white paper detailing which tasks should be handled by humans and which can be supported by AI. And you'll find a QR code at the end of our presentation to download your copy. I'm Beth Wingate, APMP Fellow and CEO of Lowfell Consulting Group. I'm joined by my colleague Brenda Christ, also an APMP Fellow and Vice President of Lowfell Consulting. Brenda holds APMP professional certification and teaches a number of Low Fell Consulting classes. We each have over thirty years of experience in bid and proposal management and writing, and we both use AI in our daily work. If you have any questions about AI in bids and proposals and how to maximize your implementation, please contact us. Let's get started with a quick audience poll, And on the next slide, you'll you'll be able to answer this. How often do you use AI tools for proposals such as finding or writing content, developing graphics, or creating presentations or marketing materials? A) We use it for every proposal B) About half the time C) Occasionally or D) I wish or our company hasn't started or won't allow us to do that. I'll give you two more seconds to answer. Let's see what we've got here. Wow, okay, so we're all over the board here now. Majority are using it for every proposal, about 21% use it half the time, some use it occasionally, and about 5% I wish. And that's changed when we've done this poll previously. The answer is there were a lot more people I wish at this point. So we are here to share some thoughts on how to develop the most effective and authentic and high scoring content using AI tools. Brenda, over to you. Tell us what AISpeak is. Thank you, Beth. Well, let me start with this presentation with a story. I teach AI, Modern Proposal Writing class at Low Volt Consulting. During this class, we ask students to write a technical response to a case study and when we give them the case study we give them background information, RFP requirements, and evaluation criteria. Then we break the class down into groups, small groups of three to four. We asked them to use their creativity and ingenuity to draft the highest scoring section. About two years ago I noticed that some people were using AI to generate their answers while others brainstormed and wrote responses and kind of strategized first. What we found after all this time is that the teams that drafted their assignment with just using AI their responses lack customer focus, responsiveness and discriminators and on the other hand the students who just you know kind of strategized first, wrote their articles, used their creativity and ingenuity were able to write in a very human way. They had great substantiation and proof points and clear strengths. So I found this to be an interesting phenomena and another caveat to that phenomena is that the teams that produce the high scoring content usually had one or more professional proposal writers in the group and the people who depended on AI were primarily SMEs without a lot of writing experience and you could definitely tell a vast difference. You know at Lowfell Consulting after seeing all these examples we coined the term AI speak. So in August 2025 I wrote an article, a blog for the Lowfell Consulting website about AI speak called Spotting AI's Faking Proposals and Why it Matters. It got more impressions and responses than any other article we wrote that year, so that was really surprising. And we had one of our one of the people who looked at our article and commented on wrote a very thoughtful response and I want to read this to you. This is exactly it. Too many proposal teams think Jenny and I will write the bid for them and what comes out is vague generic AI speak that evaluators will dismiss right away. AI can't replace hard yards of compliance checks, storyboarding, and SME input. If you can't map every sentence to a requirement or to an evaluator scoring sheet you are just filling space and losing points. GenAI is a tool not a shortcut. And that is kind of what we want to talk about for the rest of this presentation. How can we use AI effectively and responsibly to generate the highest quality proposal content? Now I would like to ask you a very sensitive question maybe. Have you ever spotted AI speak in a proposal? Yes or no? Beth, you want to give them a few minutes there before you reveal the results? You. Looks like people are starting answer. I'll give you just a couple more seconds. Sure. Okay. There you go. That's funny. That's funny. So, I'm not the only one who's noticing it. Thank you everybody. Okay, you want to go on to the next poll, Beth? Okay. Have you accidentally written AI speak yourself? Okay, guilty as charged, you know, but I guess initially when I started out, I was also guilty of writing it and not prompting as well as I could. So go ahead and put your answers in here. Give me a couple more seconds. Yeah, will say I am guilty as charged, especially initially when I, before I kind of developed a rhythm with mine. Exactly, it sounded good and I actually read what it said and went, okay. Alright. There you go. Okay, well, lots of a third of you are just better than I am, so I appreciate that. Ben you want to go on to the next slide. Here we go. Okay, so to test AI speak what I did was I wrote a simple case study. Here's the basis of my case study. I was going to request that bid and that we are requesting a bid and proposal company provide information about their tool based on the following requirements. So we wanted to start out with, functional requirements. We asked, if you can see here I have, points about the core functional requirements, automated content generation, technical requirements, data security, performance expectations, and population. So I really wanted to know about a company's AI product and these are some high level requirements that I put into the AI tool, but I also drilled down on them. So instead of under core requirements, I asked them, you know, does your tool have natural language processing? Can it understand and generate proposal content? Can it summarize RFPs and extract requirements? Then under automated content generation, you know, I said does your tool draft responses to RFP questions? You know, can it create executive summaries? Under technical requirements, I asked if it integrated with specific CRM and ERP tools and document management systems, and if it had a cloud based infrastructure and so on and so forth. So I try to get, you know, as specific as possible about my requirements and expectations for the tool. You know I put a lot more detail into it than the high level requirements that you see on the screen. Now I would like to show you what I got back for my answer. Now I am going to read the first two sentences. Our next generation AI powered proposal ecosystem revolutionizes bid the bid lifecycle through seamless orchestration of cognitive automation, contextual intelligence, and hyper collaborative cloud enablement leveraging advanced natural language processing and deep learning paradigms and platform intuitively transforms complex RP data into dynamic insight driven narratives that accelerate delivery. Beth, what do you think of that? Oh my god. Ugh! So, yeah, I think ugh! Is probably the politest thing to say. So, let us talk about why AI speak is such a problem. What's wrong with what Brenda's AI generated? I have no idea what it's trying to communicate. In fact, as Brenda and I were developing this presentation, I quickly dropped this paragraph into my private chat GBT and told it, Rewrite this into eighth grade language so I could understand what I was dealing with. Two seconds later in chat GPT, I could understand the meaning. It's about the same number of words as the previous slide, but I can actually understand them at a quick glance. Of course, both examples need metrics and proof points, but which would you rather read as an evaluator? All that gobbledygook or five or six sentences that have clear meanings. So besides being incomprehensible at a glance, what else is wrong with this AI generated proposal content? There are so many buzzwords without meaning in red. It's packed with trendy inflated terms like revolutionizes, cognitive automation, contextual intelligence, and my favorite, hyper collaborative cloud enablement. These words sound impressive, but they don't tell the reader anything about what this product actually does. In blue, there's no clear customer benefit. The passage never answers the question, why does this matter to me? Or as I like to say, so what? There's no mention of how the system helps proposal teams write faster, improve compliance, or win bids. Evaluators want outcomes, not marketing jargon. In green, there is no evidence or quantification. It makes big claims such as unparalleled agility and quantum leap in proposal velocity, but there is no data, metrics, examples to support them. Without this proof, these statements are meaningless they are what we call fluff. Four, it is overly abstract and impersonal in PRPL. It describes features like they exist in a vacuum, defining the future of intelligent proposal management. Marketing baloney. The writing is flat and bloated. Every sentence is long, formal, packed with modifiers, and the sentence length only differs by one or two words. There's no variation or conversational tone. It sounds robotic and exhausting to read aloud. And finally, there's no link to RFP requirements. It doesn't tie back to any of the customary requirements that Brenda had, acceptable quality level, or evaluation factors. A real proposal has to demonstrate compliance and strengths, and this fails to do either. So here are some other issues with AI generated text. There's a lack of depth. AI often delivers surface level summaries rather than original insight. It can restate information without analyzing causes, consequences, or context, and the results sound knowledgeable, but they are empty. There is shallow customization. AI can rarely explain why you selected your approach, how to do the job, and the inherent risks in your solution, and that involves teamwork and communication. It over explains the basics. AI sometimes assumes the reader knows nothing, leading to lengthy explanations of simple terms or concepts your audience already understands. This really slows down your writing, irritates the experts, wastes your real estate. It's got a flat tone and rhythm. It can feel monotonous, as sentences, like I said before, are often uniform in length and structure, and have very little variation or energy. It lacks the natural rhythm of our human speech and writing, making it hard to stay engaged. The language is often overly formal. AI leans towards overly polite or bureaucratic phrasing favoring business speak over authentic voice. It avoids contractions and uses filler words like in order to utilize or leverage, which is in our 100 words to avoid, or that can kill your proposals article on our website. And then finally, hallucinations. AI can present incorrect answers and information that appears authoritative at first glance, but upon closer review those responses are completely incorrect or full of baseless assumptions. I will give you some AI speak examples. The first one I often see is over explaining. This audience already knows the definition and the purpose of an RFP so the definition or explanation was not necessary. Number two, flat tone and rhythm. Each sentence is grammatically acceptable but rhythmically dull they seem to have the same cadence in their sentence and there's no flow or momentum. Know third you have overly formal language you know you can say here as you see on the screen in order to effectively leverage organizational synergies our team will utilize comprehensive framework for success. Well that is grammatically right, a human would say we will work together using a clear plan to get results. And AI can hurt your proposal scores. You know AI speak can kill your credibility. AI generated language often sounds generic, vague, and overly polished. When evaluators see phrases like interest related solutions or robust methodologies without proof they sense fluff instead of substance. The proposal loses the appearance of technical authority and realism. It can also destroy trust. Evaluators want to feel the proposal understands their requirements and that you have done similar work before. AI speak sounds kind of detached and impersonal. You can feel the writer doesn't understand the mission or the customer. The disconnect erodes trust and makes evaluators question the offer if they can perform the work or not. You can also decrease your evaluation scores. Evaluation teams base scores on evidence of strengths. AISpeak often misses these substituting vague benefits for measurable outcomes. If evaluators can't clearly link your response to the RFP requirements, they cannot assign points. Even if your company is capable of doing a fantastic job. The bottom line AI speak sounds safe and sophisticated but lacks proof, personality, and that human element that makes it hard for proposals to award your proposal the strengths that you might deserve. Okay, let us do another poll here. Here's a quick quiz to see if you can spot AI speak. Our cutting edge results driven team leverages industry best practices to deliver innovative, mission focused solutions that exceed customer ABC's AQLs. So, here's the poll. Is this AI speak? A, yes or B, no. Alright, we'll give you a couple more seconds to answer. All right, the answers are coming in. And let's hit next. Yes, this is definitely AI speak. It's just it's full of nonsense and and not Not what we're looking for here. Look at the use of buzzwords, lack of beneficial outcomes for the customer There's just new metrics. There's nothing Alright, another quiz On the next slide, we're going to click A Yes, this is AI speak or B No, and I'll let you read this here for a second. All right. Enter your answer here. I'll give you a couple more seconds to answer. Alright, no, great, good job guys. This is not AI speak, even though it uses an em dash. I was devastated when I started seeing everyone say AI loves em dashes and sticks them everywhere. I've used them for years effectively. It draws the reader's attention to a detailed resource loaded schedule, and the author substantiates its authenticity with a proof point from a similar program for that customer. So this is not AI speak. All right, last quiz. We're going to click A if this is AI speak and b if it's not AI speak. And here we go. Answer answer here. Few more seconds. And yes, this is definitely an example of AI speed. There's no evidence, no examples, or data, or results. There's no specificity. Which methodologies are we talking about? Which stakeholders? And authentic tone. It really reads like a template that anybody could write. So before we continue, I want to add that AI is a great tool and it isn't the enemy, but misuse of AI is the enemy of winning proposals. So let us stop AI speak. Here are eight ways to stop AI speak in your proposals. Number one, improve your prompt quality. Write prompts that give AI a clear role, audience, tone and date of views. Be as specific as possible. The better your input the more authentic and targeted your output. Number two, cross check your RFP responsiveness. Compare every paragraph to a solicitation requirement. If the sentence does not directly answer the customer shall, must, and will it's probably filler or fluff. Number three, eliminate filler. Complete big terms like cutting edge, best in class, or robust solution. Replace them with measurable facts, results, and examples that evaluators should score. Number four, pick the read aloud test. Read your section out loud. It sounds stiff, repetitive or robotic because AI speak right? Human's writing has rhythm and very natural way about with pauses and emotional emphasis. Number five, add human fingerprints insert real projects, success stories, lessons learned, metrics or quotes from SMEs. Authentic authentic details show ownership credibility that AI can't replace. Number six, vary your sentence structure. Make short punchy statements with longer complex ones that's kind of like the normal flow of your speech. This creates flow and energy often missing in AI. Number seven, conduct a strength based review. That means that you are going to count all the strengths for every evaluated criteria. You want to make sure that they are strong, are discriminators with your company, they are well substantiated and they will lead to high score. Number eight, you know you want to verify no hallucinations are present. I also want to add one other thing you also want to make sure that no bias is present. Do you have a certain bias about the content in your knowledge base that it might favor a certain solution or a position. You want to make sure that you are very understanding of that solution bias. So those are eight great ways to stop AI speak. So here are some use cases and tactics to improve your prompt quality. Define the role AI is playing and who your audience is. Tell the AI who it is and who it's writing for. For example, you're a senior proposal manager writing for HHS CMS federal evaluators. We're technical experts in Medicare fraud, waste, and abuse. Use professional but plain language suitable for acquisition reviewers. The details you provide will help prevent AI from defaulting to cliches. Two, support with verified data and evidence. AI writes vague text when it lacks facts, just like our human proposal writers. Feed it metrics, project results, CPARs, and customer quotes. These facts will brown the proposal text with evidence and eliminate that fluffy filler. Three, define the tone and voice you want to use in the proposal to the greatest extent possible. Point AI to specific documents or websites to give it context. The correct tone and voice will increase evaluator trust in your proposals. For example, when I'm using AI to generate text for me, I point it to our website and to our nine books as I wrote or edited much of that content. So the AI already has really good examples of my writing style. Four, keep your prompts task focused. Give the AI model an action, not just a topic. For example, summarize how our risk management approach exceeds RFP requirements tied to customer benefits. This action will keep the focus on what matters to evaluators. Five, give do not rule. Tell the AI what not to do. For example, avoid generic phrases like cutting edge, best in class, and innovative. Use clear and specific examples instead and eliminate flop. This prompt will block AI cliches before they appear. Take a look at the low felt 100 words that kill your proposals articles. I know several people that have added the 100 forbidden words into their negative guidance. I get a lot of feedback. Six, frame through strengths. Add AI to help generate a strength by giving it the formula and data needed to develop the strength. This prompt will replace vague claims with scoring content that the evaluators value. Seven, adjust and iterate for improvement. Treat prompting as a dialogue, it's not a one shot deal. If you don't like what you received, continue prompting to improve it. Your continuous feedback sharpens the writing and strips away the AI speak tone. I also use several private models. If I don't like the content one model produces, I copy my prompt over to another model and see what it can do. Sometimes I'll drop in content generated from several models and ask a model to generate content based on that previous content, and then I take that final content if I like it and rework it by hand. Eight, verbalize sampling. Use a technique from Stanford, Northeastern, and West Virginia researchers called verbalized sampling. Let's take a quick look at that on the next slide. You've got the source information. This was from a Neuron newsletter in October '5. Here's the challenge. Training AI systems to be safe and reliable often dulls their creativity, leading to repetitive, predictable answers. This happens because human reviewers of AI testing content tend to favor responses that sound conventional rather than imaginative. Why does that happen? Studies show that people naturally gravitate towards text that feels familiar and straightforward, even when more creative options might be better. That preference gets baked into the AI's training process, reducing originality over time. Verbalized sampling is one solution to this challenge. Instead of requesting one answer, you ask the AI for several variations using the magic prompt we'll see on the next slide. For example, obviously this is a simple example, give me five different jokes about dogs, and you get back much more creative responses. The university researchers suggest adding this one sentence or magic prompt before any creative task: generate five responses with their corresponding probabilities sampled from the full distribution. So instead of five nearly identical replies, you'll get five truly varied ones. For reference, I asked ChatGPT and Claude to write jokes about dogs, and many of the answers were identical between the two systems. When I used this prompt, I got completely different results. I thought the responses from Claude were more creative, but we're talking dog jokes here. This approach boosts creativity and variety in AI generated responses, roughly doubling creative quality in writing tasks without losing accuracy or reliability. You can try this technique with any AI chatbot today, though the researchers said back in October that it performed best with advanced models like ChatGPT, five, Claude, and Gemini Pro. It's essentially a way to tap into the creative potential that the AI training tends to suppress. So, try combining this technique with other techniques we discuss in this presentation and in our blog articles. And I really encourage you to read the research paper. It's really interesting. Brenda, you are on mute. So, let us explain the improved prompt which starts with a description of the role of the proctor, the objective of the prompt, the tone used to communicate with the customer. As you can see on the left hand side we have the original prompt and we talk about the major requirements that of our case study. Then in on the right hand side of the screen we got more specific and as and we got a lot more specific outside of this prop that you just see on the screen right now. So we had took great teams to define the role, objective, and the tone that we used. In addition, we had that advanced. We also given very specific instructions such as begin with a short overview paragraph, describe how the capability meets requirements, explain why it matters to proposal managers, and where possible include quantified benefits. Type each to a specific evaluation criteria and end with a summary of why it will benefit the customer. So as you can see here with these very specific instructions what we found we are going to get much more defined material and in addition to that we are going to add style rules and we are going to tell it to write in plain language, active voice, avoid cliches, each paragraph should have one strength. So, we want a very clear and beautifully written proposal. Next. So now we are going to have all of these tools that we have talked about. We are going to talk about how we can transform AI speak. Bette, you are on mute. Got to do it once. Thanks. So, as I was saying, from time to time, we are going to get AI speak from our subject matter experts. However, what steps should they take to mitigate it proactively themselves? Set writing expectations early. During your kickoff or content planning meetings, explain what good proposal writing looks like, and then it's customer focused and results driven. Provide short examples showing the difference between generic AI style text and the evidence based proposal prose you're looking for. Use structured writing templates. Structured guidance helps channel technical input into valuable content. Coach or sneeze on tone and audience. Remind them that they are writing for evaluators, not themselves. Encourage plain, confident language and short sentences over jargon or buzzwords. Request concrete evidence, not concepts. Ask your SMEs for specific examples, performance data, or results instead of abstract descriptions. Don't accept, We provide industry leading cybersecurity. Pair your SMEs with proposal writers. Assign a professional writer to each technical lead. Writers can interview the SMEs, extract insights, and then translate them into evaluator friendly text before your drafts go off the rails. Review in stages, not at the end. This is so important. Schedule short, early reviews focused only on tone clarity and compliance. Catching AI speak phrasing early avoids massive rewrites later. Provide a banned words or plain language list. Share a cheat sheet of words to avoid, like utilize, synergy, best in class, and suggest clear alternatives: use, teamwork, proven method. Take a look again at our blog 100 Words That Kill Proposals. Use a strength based review lens. When reviewing your SME input, ask does this show a strength or a proof point? Is it tied to a requirement? Can an evaluator score it? If not, revise it before it goes into a main draft. Train your on AI quality control. If SMEs use AI tools, teach them how to prompt for depth and evidence. For example, include specific metrics from our recent projects and then tell it which point it to the right projects, and how to avoid AI speak pitfalls. Reinforce the human touch. Remind your SMEs and writers that their personal insight and experience are what differentiate your proposal. Encourage short anecdotes, lessons learned, or challenges overcome. Those details win points and build trust. So, remember our example here, remember this horrendous response from the original case study here filled with buzzwords, no clear customer benefits, evidence and which quite frankly overly abstract. So let's go to the next slide. So now let's use the principles that Beth just listed in the previous slide to turn this into a high scoring proposal content. We took an AI generated content and rewrote it by hand even though you have a great prompt to start with you have to look at it, think about it, and rewrite it to be most effective. The rewrite works because it is now tied to customer requirements as you can see in green, focused on the customer in blue, names a proven methodology in orange, in this case it is our proprietary strength based solution methodology and the evidence is based on quantifiable results in purple. If I had room on this slide I would have ended with a summary of the customer benefits. It is also written in an authentic tone and sounds like a knowledgeable professional not a marketing bot. Bottom line, it is all about using the AI tools correctly and creatively. Get some solid techniques and tactics pulled together that help you in your daily content development and continue to read about and experiment with new techniques for AI content generation, keeping in mind that the goal is to develop authentic sounding, evidence based proposal text tied to the RFP requirements with quantifiable results that focus on your customer. As bid proposal professionals and business development professionals of course, let's use AI responsibly. We are offering you a complimentary white paper describing which tasks you should control and which tasks AI can support. So if you take out your cameras right now you can take a picture of a QR code and download that checklist for free. Low Fell Consulting also teaches classes in AI for proposal professionals and modern proposal writing that provides hands on training related to how AI can be used responsibly to you know as a force multiplier to drive the efficiency of your proposals and save B and P dollars. So, is quite a lot of benefits there. Beth? Thanks Brenda. Okay, over to the Deltek folks. Thank you so much Beth and Brenda, that was incredible. Before we conclude, we want to remind you that you will receive an on demand recording of today's webinar via email within twenty four hours. If we were not able to get to your questions today, we will be sure to follow-up with you offline. Also, if you could please fill out a short survey that you'll see at the conclusion of the webinar, we would greatly appreciate it. Thank you and have a great day. Please visit delltech.com for more valuable Deltek events.